docs: add 0.2.10 changelog

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
fix: audit — score falsy-zero, crop person mismatch, adaptive cap, suppress_output, ldconfig glob, scheduler sleep
2026-06-12 16:51:53 +00:00 · 2026-06-12 16:48:24 +00:00 · 2026-06-12 16:38:40 +00:00 · 2026-06-12 15:58:29 +00:00 · 2026-06-12 15:50:51 +00:00 · 2026-06-12 15:49:28 +00:00
26 changed files with 2607 additions and 424 deletions
@@ -4,7 +4,10 @@ API_KEY=your-immich-api-key
 FRIGATE_URL=http://192.168.1.10:5000
 # ── Mode & Strategy ───────────────────────────────────────────────────────────
-AUTO_MODE=true
+# Auto mode is active by default when no TTY is present (Docker/cron).
 # Set AUTO_MODE=true to force auto mode even in an interactive terminal.
 # AUTO_MODE=true
 # VERBOSE=true                # Enable DEBUG-level console output (log file is always DEBUG)
 # TRAINING_MODE: face = upload to Frigate face recognition API
 #                object = save crops to output dir for manual Frigate placement
 TRAINING_MODE=face
@@ -29,8 +32,8 @@ STRATEGY=auto
 # MAX_AUTO_IMAGES=80          # Hard cap on auto-diversity selection (default: 80)
 # ── Caching & Models ──────────────────────────────────────────────────────────
-FORCE_CPU=false
+# FORCE_CPU=true              # Disable GPU, fall back to CPU
-ENABLE_CACHE=true
+# ENABLE_CACHE=false          # Disable embedding cache (default: true)
 CACHE_DIR=/app/.if_cache
 HF_HOME=/models/huggingface
 INSIGHTFACE_HOME=/models/.insightface
@@ -41,5 +44,10 @@ INSIGHTFACE_HOME=/models/.insightface
 # RESET_PERSON=John           # Clear uploaded+rejected history for one person
 # ── Scheduling ────────────────────────────────────────────────────────────────
-# Cron expression (unset = run once and exit)
+# CRON_SCHEDULE controls container lifetime:
-CRON_SCHEDULE=0 3 * * 0     # Every Sunday at 3 AM
+#   unset          — run once on startup, then exit
 #   empty string   — stay alive, run nothing (trigger manually: docker exec -it winnow winnow)
 #   cron expression — run on startup, then on schedule
 # CRON_SCHEDULE=              # Manual mode (keep alive, no auto-run)
 # CRON_SCHEDULE=0 3 * * 0    # Every Sunday at 3 AM
 # CRON_SCHEDULE=0 3 1 * *    # First of every month
@@ -126,3 +126,58 @@ jobs:
      - name: Inspect image
        run: |
          docker buildx imagetools inspect ${{ steps.tags.outputs.tags }}
  build-cpu:
    name: Build CPU-only (amd64)
    runs-on: ubuntu-latest
    permissions:
      contents: read
      packages: write
    steps:
      - name: Free up disk space
        run: |
          sudo rm -rf /usr/share/dotnet
          sudo rm -rf /opt/ghc
          sudo rm -rf "/usr/local/share/boost"
          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
          echo "Disk space freed."
      - name: Checkout repository
        uses: actions/checkout@v6
      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@v4
      - name: Log in to GHCR
        uses: docker/login-action@v4
        with:
          registry: ${{ env.REGISTRY }}
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}
      - name: Determine CPU image tag
        id: cpu-tag
        run: |
          if [ "${{ github.ref_name }}" = "dev" ]; then
            echo "tag=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:dev-cpu" >> "$GITHUB_OUTPUT"
          else
            echo "tag=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:cpu" >> "$GITHUB_OUTPUT"
          fi
      - name: Build and push CPU image
        uses: docker/build-push-action@v6
        with:
          context: .
          file: ./Dockerfile
          platforms: linux/amd64
          build-args: VARIANT=cpu
          cache-from: type=gha,scope=linux/amd64-cpu
          cache-to: type=gha,mode=max,scope=linux/amd64-cpu
          github-token: ${{ secrets.GITHUB_TOKEN }}
          push: true
          tags: ${{ steps.cpu-tag.outputs.tag }}
      - name: Inspect CPU image
        run: |
          docker buildx imagetools inspect ${{ steps.cpu-tag.outputs.tag }}
@@ -1,239 +1,36 @@
-# Custom
+# Output and runtime artefacts
 frigate_train/
 *.log
 runs/
 # Model and cache files
 yolov9c.pt
 .insightface/
 .huggingface/
 .cache/huggingface
 .if_cache/
 .immich_config.json
 # Python-generated files
 __pycache__/
 *.py[oc]
 build/
 dist/
 wheels/
 *.egg-info
-# Virtual environments
+# Secrets
 .venv
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[codz]
 *$py.class
 # C extensions
 *.so
 # Distribution / packaging
 .Python
 build/
 develop-eggs/
 dist/
 downloads/
 eggs/
 .eggs/
 lib/
 lib64/
 parts/
 sdist/
 var/
 wheels/
 share/python-wheels/
 *.egg-info/
 .installed.cfg
 *.egg
 MANIFEST
 # PyInstaller
 #   Usually these files are written by a python script from a template
 #   before PyInstaller builds the exe, so as to inject date/other infos into it.
 *.manifest
 *.spec
 # Installer logs
 pip-log.txt
 pip-delete-this-directory.txt
 # Unit test / coverage reports
 htmlcov/
 .tox/
 .nox/
 .coverage
 .coverage.*
 .cache
 nosetests.xml
 coverage.xml
 *.cover
 *.py.cover
 .hypothesis/
 .pytest_cache/
 cover/
 # Translations
 *.mo
 *.pot
 # Django stuff:
 *.log
 local_settings.py
 db.sqlite3
 db.sqlite3-journal
 # Flask stuff:
 instance/
 .webassets-cache
 # Scrapy stuff:
 .scrapy
 # Sphinx documentation
 docs/_build/
 # PyBuilder
 .pybuilder/
 target/
 # Jupyter Notebook
 .ipynb_checkpoints
 # IPython
 profile_default/
 ipython_config.py
 # pyenv
 #   For a library or package, you might want to ignore these files since the code is
 #   intended to run in multiple environments; otherwise, check them in:
 # .python-version
 # pipenv
 #   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
 #   However, in case of collaboration, if having platform-specific dependencies or dependencies
 #   having no cross-platform support, pipenv may install dependencies that don't work, or not
 #   install all needed dependencies.
 # Pipfile.lock
 # UV
 #   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
 #   This is especially recommended for binary packages to ensure reproducibility, and is more
 #   commonly ignored for libraries.
 # uv.lock
 # poetry
 #   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
 #   This is especially recommended for binary packages to ensure reproducibility, and is more
 #   commonly ignored for libraries.
 #   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
 # poetry.lock
 # poetry.toml
 # pdm
 #   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
 #   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
 #   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
 # pdm.lock
 # pdm.toml
 .pdm-python
 .pdm-build/
 # pixi
 #   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
 # pixi.lock
 #   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
 #   in the .venv directory. It is recommended not to include this directory in version control.
 .pixi
 # PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
 __pypackages__/
 # Celery stuff
 celerybeat-schedule
 celerybeat.pid
 # Redis
 *.rdb
 *.aof
 *.pid
 # RabbitMQ
 mnesia/
 rabbitmq/
 rabbitmq-data/
 # ActiveMQ
 activemq-data/
 # SageMath parsed files
 *.sage.py
 # Environments
 .env
 .envrc
 .venv
 env/
 venv/
 ENV/
 env.bak/
 venv.bak/
-# Spyder project settings
+# Python
-.spyderproject
+__pycache__/
-.spyproject
+*.py[oc]
 *.so
 .Python
-# Rope project settings
+# Packaging
-.ropeproject
+build/
 dist/
 *.egg-info/
 wheels/
-# mkdocs documentation
+# Virtual environments
-/site
+.venv/
-# mypy
+# Tools
 .mypy_cache/
 .dmypy.json
 dmypy.json
 # Pyre type checker
 .pyre/
 # pytype static type analyzer
 .pytype/
 # Cython debug symbols
 cython_debug/
 # PyCharm
 #   JetBrains specific template is maintained in a separate JetBrains.gitignore that can
 #   be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
 #   and can be added to the global gitignore or merged into this file.  For a more nuclear
 #   option (not recommended) you can uncomment the following to ignore the entire idea folder.
 # .idea/
 # Abstra
 #   Abstra is an AI-powered process automation framework.
 #   Ignore directories containing user credentials, local state, and settings.
 #   Learn more at https://abstra.io/docs
 .abstra/
 # Visual Studio Code
 #   Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
 #   that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
 #   and can be added to the global gitignore or merged into this file. However, if you prefer, 
 #   you could uncomment the following to ignore the entire vscode folder
 # .vscode/
 # Ruff stuff:
 .ruff_cache/
-
+.pytest_cache/
-# PyPI configuration file
+.mypy_cache/
-.pypirc
+.python-version
 # Marimo
 marimo/_static/
 marimo/_lsp/
 __marimo__/
 # Streamlit
 .streamlit/secrets.toml
 compose.override.yml
@@ -1 +0,0 @@
 3.13
@@ -7,6 +7,50 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 ## [0.2.10] - 2026-06-12
 ### Added
 - **`VERBOSE` env var**: set `VERBOSE=true` to enable DEBUG-level console output. The log file always captures DEBUG; this flag controls what appears on the terminal. Useful when diagnosing issues without a full shell into the container.
 - **`:cpu` Docker image tag**: a separate CPU-only image (`ghcr.io/sudolulo/winnow:cpu`) is now built and pushed alongside `:latest`. Uses `onnxruntime` instead of `onnxruntime-gpu`; ~2 GB smaller. Suitable for systems without an NVIDIA GPU.
 - **Empty `CRON_SCHEDULE` keeps container alive**: setting `CRON_SCHEDULE=` (empty string) starts the container without running immediately and without exiting — useful for `docker exec` ad-hoc runs on a long-lived container. Previously, an empty value was treated the same as unset (run once, then exit).
 ### Changed
 - **TTY auto-detection replaces `AUTO_MODE`**: winnow now detects whether a TTY is attached (`sys.stdin.isatty()`) and switches between interactive and auto mode automatically. `AUTO_MODE=true` becomes an explicit override for forcing auto mode in a terminal session. No config change needed for normal Docker deployments.
 - **Logging levels audited**: internal algorithmic detail (clustering steps, asset fetch progress, per-page pagination) demoted from INFO to DEBUG. INFO now reflects meaningful pipeline milestones only (model ready, selection complete, quality filtered). Reduces noise in production logs without losing information.
 - **Model load logging improved**: SigLIP and InsightFace loading now reports cache hit/miss, download size estimate, device used, and load time.
 ### Fixed
 - **GPU OOM crash loop (production)**: `CUDAExecutionProvider` was silently absent even with a GPU attached, causing InsightFace to run on CPU and exhaust RAM processing large person libraries. Root cause: CUDA/cuDNN libraries in nvidia pip packages were invisible to onnxruntime. Fixed by running `ldconfig` over all `nvidia-*/lib/` directories in the venv at image build time.
 - **ldconfig path now Python-version-agnostic**: the `find` command used to register nvidia pip libraries hardcoded `python3.13`; replaced with `python3.*` glob so the path survives a Python upgrade without silently producing an empty ldconfig config.
 - **`PYTHONPATH=/app` added to Dockerfile**: the entry point script sets `sys.path[0]` to the script directory, not `/app`. Since `uv sync` runs before `COPY winnow/`, the wheel has only dist-info in site-packages. `PYTHONPATH=/app` makes the `winnow` package importable without reverting to `python -m`.
 - **CPU fallback retrying broken GPU provider**: InsightFace CPU fallback omitted `providers=["CPUExecutionProvider"]`, causing onnxruntime to retry `CUDAExecutionProvider` on every inference call. Now explicitly sets the CPU provider and suppresses C-extension noise via fd-level redirect.
 - **`_suppress_output` stderr loss on fd exhaustion**: if the first `os.dup2` in the finally block raised `OSError`, the second call was skipped, permanently redirecting stderr to `/dev/null` for the process lifetime. Wrapped in nested `try/finally` so both restores are always attempted.
 - **Frigate `/api/faces` response parsing**: the response is `{person_name: [files], "train": [...]}` — `"train"` is a flat pending list, not a person. Previous code called `.items()` on the `"train"` value (a list), crashing with `AttributeError`. Now skips the `"train"` key explicitly.
 - **Immich 401 detection**: a stale or invalid API key now logs a clear error message (`Immich API key is invalid or expired (401 Unauthorized)`) instead of raising an unhandled exception.
 - **Falsy-zero detection confidence**: `face.get("score") or face.get("confidence")` treated a valid `score=0.0` as falsy, falling through to the `confidence` field (often `None`). Replaced with an explicit `None` check. Affected both quality filtering and hard-example weighting in diversity selection.
 - **Face crop using wrong person's image dimensions**: in multi-person assets, `_crop_face_from_thumbnail`'s scale-factor loop matched the first person with any face regardless of `person_id`, producing incorrectly scaled bounding box coordinates for the target person. Loop now applies the same `person_id` filter as `_get_face_bbox`.
 - **Adaptive stopping bypassed for partially-trained people**: in auto mode with `already_uploaded > 0`, `limit` was converted from `"auto"` to an integer, disabling the FPS adaptive threshold and early-stop check. Now keeps `limit="auto"` through selection and trims the result to the remaining capacity afterward.
 - **Embedding cache key mismatch**: HuggingFace cache path check hardcoded the model slug string; replaced with a derivation from `model_name` using `"models--" + model_name.replace("/", "--")` so the check stays correct if the model name changes.
 - **Scheduler: sleep until next run**: the loop slept a fixed 60 seconds regardless of schedule interval, causing runs to fire up to 59 seconds late and waking the process unnecessarily on long schedules (e.g. weekly). Now sleeps exactly until `next_run`.
 - **Scheduler swallowing `SystemExit`**: `except BaseException` in the run wrapper was replaced with `except Exception` (with `KeyboardInterrupt` re-raised above), so `sys.exit()` calls propagate correctly.
 - **Log handler leak**: `setup_logging` now closes and removes existing handlers before adding new ones, preventing file handle accumulation across repeated calls.
 - **`RETRY_REJECTED` silently applied in interactive mode**: the env var was applied unconditionally even in interactive sessions. Now used only as the default for the interactive prompt so users can override it per-run.
 - **`compose.yml` comment inverted**: a comment stated `-it` forces non-interactive mode; corrected to reflect that `-it` allocates a TTY (interactive mode).
 ### Security
 - API key is no longer stored in any config file. All authentication uses environment variables or `.env` only.
 ## [0.2.9] - 2026-06-12
 ### Fixed
 - **arm64: `onnxruntime-gpu` has no arm64 wheels**: `onnxruntime-gpu` only publishes `manylinux_2_27_x86_64` and `manylinux_2_28_x86_64` wheels — `uv sync` on arm64 failed with exit code 2. Gated `onnxruntime-gpu` behind `sys_platform == 'linux' and platform_machine == 'x86_64'`; arm64 and non-Linux installs now get the CPU `onnxruntime` package instead.
 - **uv lockfile now covers arm64**: Added `required-environments` to `[tool.uv]` so the lockfile is solved for both `linux/x86_64` and `linux/aarch64`, preventing silent resolution gaps for the non-build platform.
 ## [0.2.8] - 2026-06-12
 ### Fixed
@@ -1,20 +1,25 @@
-# ── Platform-conditional base ─────────────────────────────────────────────
+# ── Base images ───────────────────────────────────────────────────────────────
-#   amd64:  NVIDIA CUDA 13.3 (GPU acceleration when available, CPU fallback)
+#   amd64 + gpu:  NVIDIA CUDA 13.3 + cuDNN (GPU acceleration when available)
-#   arm64:  Ubuntu 24.04 (CPU-only; no CUDA on ARM)
+#   amd64 + cpu:  Ubuntu 22.04 (CPU-only, ~2 GB smaller image)
 #   arm64:        Ubuntu 24.04 (CPU-only; no CUDA wheels on ARM)
-FROM --platform=$BUILDPLATFORM nvidia/cuda:13.3.0-cudnn-runtime-ubuntu22.04 AS base-amd64
+ARG VARIANT=gpu
 FROM ubuntu:24.04 AS base-arm64
-# ── Build stage ───────────────────────────────────────────────────────────
+FROM --platform=$BUILDPLATFORM nvidia/cuda:13.3.0-cudnn-runtime-ubuntu22.04 AS base-amd64-gpu
 FROM ubuntu:22.04 AS base-amd64-cpu
 FROM ubuntu:24.04 AS base-arm64-gpu
 FROM ubuntu:24.04 AS base-arm64-cpu
 # ── Build stage ───────────────────────────────────────────────────────────────
 ARG TARGETARCH
-FROM base-${TARGETARCH} AS build
+FROM base-${TARGETARCH}-${VARIANT} AS build
 ARG VARIANT=gpu
 ENV DEBIAN_FRONTEND=noninteractive
-# Both bases (Ubuntu 22.04 CUDA / Ubuntu 24.04) need Python 3.13 from the
+# Both Ubuntu 22.04 and 24.04 get Python 3.13 from the deadsnakes PPA.
-# deadsnakes PPA. GNUPGHOME is isolated to a tmpdir so gpg never tries to
+# GNUPGHOME is isolated so gpg never contacts an agent socket under QEMU.
 # contact an agent socket, which fails silently under QEMU.
 RUN apt-get update && apt-get install -y --no-install-recommends \
    ca-certificates curl gnupg software-properties-common \
    && GNUPGHOME=$(mktemp -d) add-apt-repository ppa:deadsnakes/ppa -y \
@@ -30,20 +35,26 @@ RUN curl -LsSf https://astral.sh/uv/install.sh | sh \
 WORKDIR /app
-COPY pyproject.toml uv.lock ./
+# For cpu variant, swap in the CPU-only pyproject and lockfile before syncing.
-RUN uv sync --frozen --no-dev \
+COPY pyproject.toml uv.lock pyproject-cpu.toml uv-cpu.lock ./
 RUN if [ "$VARIANT" = "cpu" ]; then \
        cp pyproject-cpu.toml pyproject.toml && \
        cp uv-cpu.lock uv.lock; \
    fi && \
    uv sync --frozen --no-dev \
    && uv cache clean
 COPY winnow/ winnow/
 COPY entrypoint.sh scheduler.py ./
 RUN chmod +x /app/entrypoint.sh
-# ── Runtime stage ─────────────────────────────────────────────────────────
+# ── Runtime stage ─────────────────────────────────────────────────────────────
 #   Starts fresh from the base image — excludes build tools (g++,
 #   python3.13-dev, gnupg, software-properties-common) not needed at runtime.
-FROM base-${TARGETARCH} AS runtime
+FROM base-${TARGETARCH}-${VARIANT} AS runtime
 ARG VARIANT=gpu
 ENV DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && apt-get install -y --no-install-recommends \
@@ -61,9 +72,11 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 COPY --from=build /app /app
 COPY --from=build /usr/local/bin/uv /usr/local/bin/uv
-# Expose CUDA/cuDNN libraries from pip packages so onnxruntime-gpu
+# Register every nvidia pip-package lib/ directory with ldconfig so that
-# can find libcublasLt.so.12 and libcudnn.so.9 at runtime (amd64 only)
+# onnxruntime-gpu and torch can find libcudnn, libcublas, libcufft, etc.
-ENV LD_LIBRARY_PATH="/app/.venv/lib/python3.13/site-packages/nvidia/cudnn/lib:/app/.venv/lib/python3.13/site-packages/nvidia/cuda_runtime/lib:${LD_LIBRARY_PATH}"
+# without a hand-maintained LD_LIBRARY_PATH. Skipped silently on cpu builds.
 RUN find /app/.venv/lib/python3.*/site-packages/nvidia -type d -name "lib" \
        2>/dev/null > /etc/ld.so.conf.d/nvidia-pip.conf && ldconfig || true
 RUN groupadd -g 568 apps && useradd -u 568 -g apps -m -s /bin/bash appuser \
    && mkdir -p /models/.insightface /models/huggingface \
@@ -71,7 +84,10 @@ RUN groupadd -g 568 apps && useradd -u 568 -g apps -m -s /bin/bash appuser \
 WORKDIR /app
 USER appuser
-ENV HF_HOME=/models/huggingface INSIGHTFACE_HOME=/models
+# PYTHONPATH=/app makes the winnow package importable from the entry point script.
 # uv sync builds the wheel before winnow/ is COPY'd, so site-packages has only
 # the dist-info. Explicitly adding /app lets Python find winnow/__init__.py there.
 ENV HF_HOME=/models/huggingface INSIGHTFACE_HOME=/models/.insightface PYTHONPATH=/app
 HEALTHCHECK CMD test -f /app/entrypoint.sh || exit 1
 ENTRYPOINT ["tini", "--", "/app/entrypoint.sh"]
@@ -111,17 +111,23 @@ If the embedding model is unavailable, the tool falls back to **time spread**: e
 ## Running in Docker
 ### Image Tags
 | Tag | Arch | GPU | Notes |
 | :-- | :-- | :-- | :-- |
 | `:latest` | amd64 + arm64 | CUDA 13.3 (amd64) | Requires NVIDIA Container Toolkit on amd64 |
 | `:cpu` | amd64 | None | ~2 GB smaller; use if you have no NVIDIA GPU |
 ### Quick Start
 ```yaml
 services:
  winnow:
-    image: ghcr.io/sudolulo/winnow:latest
+    image: ghcr.io/sudolulo/winnow:latest   # or :cpu for CPU-only amd64
    environment:
      - IMMICH_URL=http://192.168.1.10:2283
      - API_KEY=your-immich-api-key
      - FRIGATE_URL=http://192.168.1.10:5000
      - AUTO_MODE=true
      - CRON_SCHEDULE=0 3 * * 0   # Every Sunday at 3 AM
    volumes:
      - /path/to/models:/models
@@ -136,11 +142,21 @@ services:
              capabilities: [gpu]
 ```
 > **CPU users (`:cpu` tag):** remove the `deploy.resources` block — no NVIDIA runtime needed.
 See [compose.yml](compose.yml) for the full annotated example.
 ### Scheduling Behaviour
-On startup the container always runs once immediately. If `CRON_SCHEDULE` is set, it then starts a scheduler that fires on the defined interval, keeping the process (and loaded models) alive between runs. Without `CRON_SCHEDULE` the container exits after the first run.
+`CRON_SCHEDULE` controls container lifetime:
 | `CRON_SCHEDULE` value | Behaviour |
 | :-- | :-- |
 | *(unset)* | Run once on startup, then exit |
 | *(empty string)* | Stay alive, run nothing — trigger manually with `docker exec -it winnow winnow` |
 | Cron expression | Run on startup, then repeat on schedule |
 In scheduled mode the process (and loaded models) stays resident between runs. In manual mode the container idles indefinitely with `sleep infinity` — useful when you want to trigger runs interactively on demand without pulling a new container each time.
 The first run after a fresh install downloads the embedding models (~1-2 GB). Subsequent runs use the cached models from the mounted volume and start immediately.
@@ -152,7 +168,8 @@ The first run after a fresh install downloads the embedding models (~1-2 GB). Su
 | Variable | Default | Description |
 | :--- | :--- | :--- |
-| `AUTO_MODE` | `false` | Run without interactive prompts — required for Docker/cron use |
+| `AUTO_MODE` | *(auto)* | Force non-interactive mode even in a terminal; auto-detected otherwise (no TTY = auto) |
 | `VERBOSE` | `false` | Set to `true` to enable DEBUG-level console output (the log file is always DEBUG) |
 | `TRAINING_MODE` | `face` | `face` — upload crops to Frigate API; `object` — save crops to disk |
 | `STRATEGY` | `auto` | `auto` (adaptive), `standard` (30 images), `broad` (100 images) |
 | `LIMIT` | *(unset)* | Exact image count — overrides `STRATEGY` |
@@ -192,7 +209,7 @@ The first run after a fresh install downloads the embedding models (~1-2 GB). Su
 | Variable | Default | Description |
 | :--- | :--- | :--- |
 | `FORCE_CPU` | `false` | Disable GPU — fall back to CPU for embedding computation |
-| `ENABLE_CACHE` | `false` | Cache computed embeddings to disk (speeds up re-runs on the same library) |
+| `ENABLE_CACHE` | `true` | Cache computed embeddings to disk (speeds up re-runs on the same library) |
 | `CACHE_DIR` | `.if_cache` | Path for embedding cache and upload tracker files |
 | `HF_HOME` | *(system)* | HuggingFace model cache location (SigLIP) |
 | `INSIGHTFACE_HOME` | *(system)* | InsightFace model cache location (Buffalo_L) |
@@ -209,7 +226,7 @@ The first run after a fresh install downloads the embedding models (~1-2 GB). Su
 | Variable | Default | Description |
 | :--- | :--- | :--- |
-| `CRON_SCHEDULE` | *(unset)* | Cron expression for recurring runs — unset exits after first run |
+| `CRON_SCHEDULE` | *(unset)* | Unset = run once and exit; empty = stay alive for manual `docker exec`; cron expression = scheduled |
 ---
@@ -222,7 +239,7 @@ uv sync
 uv run winnow
 ```
-Requires Python 3.12+ and [uv](https://astral.sh/uv/). An NVIDIA GPU is strongly recommended — CPU mode works but embedding computation is significantly slower.
+Requires Python 3.13+ and [uv](https://astral.sh/uv/). An NVIDIA GPU is strongly recommended — CPU mode works but embedding computation is significantly slower.
 ---
@@ -231,7 +248,7 @@ Requires Python 3.12+ and [uv](https://astral.sh/uv/). An NVIDIA GPU is strongly
 - **Immich** v1.106+
 - **Frigate** v0.16+ (face mode only — object mode has no Frigate API dependency)
 - **NVIDIA GPU** recommended (CUDA 12.x)
- **Python 3.12+**
+- **Python 3.13+**
 ---
@@ -9,7 +9,10 @@ services:
      - FRIGATE_URL=${FRIGATE_URL}
      # ── Mode & Strategy ───────────────────────────────────────────────────
-      - AUTO_MODE=true
+      # Auto mode is active by default when no TTY is present (Docker/cron).
      # Set AUTO_MODE=true to force auto mode in an interactive terminal.
      # To run interactively: docker exec -it winnow winnow
      # - VERBOSE=true                # Enable DEBUG-level console output
      # TRAINING_MODE: face = upload to Frigate face recognition API
      #                object = save crops to output dir for manual Frigate placement
      - TRAINING_MODE=face
@@ -34,8 +37,8 @@ services:
      # - MAX_AUTO_IMAGES=80          # Hard cap on auto-diversity selection (default: 80)
      # ── Caching & Models ──────────────────────────────────────────────────
-      - FORCE_CPU=false
+      # - FORCE_CPU=true              # Disable GPU, fall back to CPU
-      - ENABLE_CACHE=true
+      # - ENABLE_CACHE=false          # Disable embedding cache (default: true)
      - CACHE_DIR=/app/.if_cache
      - HF_HOME=/models/huggingface
      - INSIGHTFACE_HOME=/models/.insightface
@@ -46,18 +49,19 @@ services:
      # - RESET_PERSON=John           # Clear uploaded+rejected history for one person
      # ── Scheduling ────────────────────────────────────────────────────────
-      # Cron expression (unset = run once and exit)
+      # CRON_SCHEDULE controls container lifetime:
-      # Every Sunday at 3 AM:
+      #   unset          — run once on startup, then exit
-      - CRON_SCHEDULE=0 3 * * 0
+      #   empty string   — stay alive, run nothing; trigger manually with:
-      # - CRON_SCHEDULE=0 3 1 * *
+      #                    docker exec -it winnow winnow
-      # - CRON_SCHEDULE=*/30 * * * *
+      #   cron expression — run on startup, then on schedule
      # - CRON_SCHEDULE=                # Manual mode (keep alive, no auto-run)
      # - CRON_SCHEDULE=0 3 * * 0      # Every Sunday at 3 AM
      # - CRON_SCHEDULE=0 3 1 * *      # First of every month
    volumes:
      # Replace with absolute paths on your host, e.g. /opt/winnow/models
      - /path/to/winnow/models:/models
      - /path/to/winnow/cache:/app/.if_cache
      - /path/to/winnow/output:/app/frigate_train
    stdin_open: true
    tty: true
    restart: unless-stopped
    deploy:
      resources:
@@ -2,11 +2,19 @@
 set -e
 export PYTHONUNBUFFERED=1
-# 1. Run the job immediately on startup
+# CRON_SCHEDULE controls container lifetime:
 #   unset          — run once and exit
 #   empty string   — stay alive, run nothing (use: docker exec -it winnow winnow)
 #   cron expression — run immediately, then on schedule
 if [ "${CRON_SCHEDULE+isset}" = "isset" ] && [ -z "$CRON_SCHEDULE" ]; then
    echo "▶ CRON_SCHEDULE is empty — manual mode. Use 'docker exec -it winnow winnow' to run."
    exec sleep infinity
 fi
 echo "▶ Running on startup..."
-/app/.venv/bin/python -m winnow.cli
+/app/.venv/bin/winnow
 # 2. If a schedule exists, start the scheduler
 if [ -n "${CRON_SCHEDULE:-}" ]; then
    echo "▶ CRON_SCHEDULE set to: $CRON_SCHEDULE"
    echo "▶ Switching to scheduled mode..."
@@ -14,4 +22,3 @@ if [ -n "${CRON_SCHEDULE:-}" ]; then
 else
    echo "▶ No schedule set, exiting."
 fi
@@ -0,0 +1,93 @@
 [project]
 name = "winnow"
 version = "0.2.10"
 description = "Immich to Frigate training sets"
 license = "AGPL-3.0-or-later"
 requires-python = ">=3.13"
 authors = [{ name = "Holden Salomon", email = "holden@arch.fyi" }]
 keywords = ["immich", "frigate", "face-recognition", "training-data", "arcface", "insightface"]
 classifiers = [
    "Development Status :: 3 - Alpha",
    "Intended Audience :: Developers",
    "License :: OSI Approved :: GNU Affero General Public License v3 or later (AGPLv3+)",
    "Programming Language :: Python :: 3.13",
    "Topic :: Scientific/Engineering :: Image Recognition",
 ]
 dependencies = [
    "croniter>=5.0.2",
    "insightface>=0.7.3",
    "numpy>=2.2.6",
    "onnxruntime>=1.23.2",
    "opencv-python-headless>=4.12.0.88",
    "pillow>=12.1.0",
    "python-dotenv>=1.2.1",
    "requests>=2.32.5",
    "rich>=14.2.0",
    "torch>=2.12.0",
    "torchvision>=0.27.0",
    "transformers>=4.57.6",
    "ultralytics>=8.4.66",
 ]
 [project.scripts]
 winnow = "winnow.cli:main"
 [project.urls]
 Repository = "https://github.com/sudolulo/winnow"
 [tool.uv]
 required-environments = [
    "sys_platform == 'linux' and platform_machine == 'x86_64'",
 ]
 [tool.uv.sources]
 torch = [
    { index = "pytorch-cpu", marker = "sys_platform == 'linux' and platform_machine == 'x86_64'" },
 ]
 torchvision = [
    { index = "pytorch-cpu", marker = "sys_platform == 'linux' and platform_machine == 'x86_64'" },
 ]
 [[tool.uv.index]]
 name = "pytorch-cpu"
 url = "https://download.pytorch.org/whl/cpu"
 explicit = true
 [dependency-groups]
 dev = [
    "pytest>=8.0",
    "ruff>=0.15.17",
 ]
 [tool.hatch.build.targets.wheel]
 packages = ["winnow"]
 [tool.ruff]
 line-length = 120
 target-version = "py313"
 [tool.ruff.lint]
 select = ["E", "F", "I"]
 [tool.deptry]
 pep621_dev_dependency_groups = ["dev"]
 [tool.deptry.package_module_name_map]
 pillow = "PIL"
 opencv-python-headless = "cv2"
 python-dotenv = "dotenv"
 insightface = "insightface"
 numpy = "numpy"
 onnxruntime = "onnxruntime"
 requests = "requests"
 rich = "rich"
 torch = "torch"
 transformers = "transformers"
 ultralytics = "ultralytics"
 [tool.pytest.ini_options]
 testpaths = ["tests"]
 [build-system]
 requires = ["hatchling"]
 build-backend = "hatchling.build"
@@ -1,6 +1,6 @@
 [project]
 name = "winnow"
-version = "0.2.8"
+version = "0.2.10"
 description = "Immich to Frigate training sets"
 license = "AGPL-3.0-or-later"
 requires-python = ">=3.13"
@@ -18,7 +18,9 @@ dependencies = [
    "insightface>=0.7.3",
    "nvidia-cudnn-cu12>=9.0.0",
    "numpy>=2.2.6",
-    "onnxruntime-gpu>=1.23.2",
+    "onnxruntime-gpu>=1.23.2; sys_platform == 'linux' and platform_machine == 'x86_64'",
    "onnxruntime>=1.23.2; sys_platform == 'linux' and platform_machine != 'x86_64'",
    "onnxruntime>=1.23.2; sys_platform != 'linux'",
    "opencv-python-headless>=4.12.0.88",
    "pillow>=12.1.0",
    "python-dotenv>=1.2.1",
@@ -37,7 +39,11 @@ winnow = "winnow.cli:main"
 Repository = "https://github.com/sudolulo/winnow"
 [tool.uv]
-override-dependencies = ["onnxruntime-gpu>=1.23.2"]
+override-dependencies = ["onnxruntime>=1.23.2"]
 required-environments = [
    "sys_platform == 'linux' and platform_machine == 'x86_64'",
    "sys_platform == 'linux' and platform_machine == 'aarch64'",
 ]
 [tool.uv.sources]
 torch = [
@@ -1,7 +1,6 @@
 #!/usr/bin/env python3
 import logging
 import os
 import subprocess
 import sys
 import time
 from pathlib import Path
@@ -9,33 +8,27 @@ from pathlib import Path
 try:
    from croniter import croniter
 except ImportError:
-    print("❌ croniter not installed. Run: uv add croniter")
+    print("croniter not installed. Run: uv add croniter")
    sys.exit(1)
 # Imported at module level so models loaded during the first run stay
 # resident in memory across all subsequent scheduled runs.
 from winnow.cli import main
 SCHEDULE = os.environ["CRON_SCHEDULE"]
 MODELS_DIR = os.environ.get("HF_HOME", "/models/huggingface")
-INSIGHTFACE_BASE = os.environ.get("INSIGHTFACE_HOME", "/models")
+INSIGHTFACE_HOME = os.environ.get("INSIGHTFACE_HOME", "/models/.insightface")
 RUN_ENV = {**os.environ, "PYTHONUNBUFFERED": "1"}
 logger = logging.getLogger(__name__)
-def check_models():
+def check_models() -> None:
-    """Log model status before each run."""
+    buffalo = Path(INSIGHTFACE_HOME) / "models" / "buffalo_l"
    print("📦 Checking models...", flush=True)
    buffalo = Path(INSIGHTFACE_BASE) / ".insightface" / "models" / "buffalo_l"
    if buffalo.exists():
        print("  ✅ InsightFace Buffalo_L: present", flush=True)
    else:
        print("  ⬇️  InsightFace Buffalo_L: not found — will download", flush=True)
    hf_hub = Path(MODELS_DIR) / "hub"
-    if hf_hub.exists() and any(hf_hub.iterdir()):
+    if not buffalo.exists():
-        print("  ✅ HuggingFace models: present", flush=True)
+        print("  InsightFace Buffalo_L not found — will download on first run", flush=True)
-    else:
+    if not (hf_hub.exists() and any(hf_hub.iterdir())):
-        print("  ⬇️  HuggingFace models: not found — will download", flush=True)
+        print("  HuggingFace models not found — will download on first run", flush=True)
    print("🚀 Starting winnow...", flush=True)
 NOW = time.time()
@@ -45,14 +38,15 @@ next_run = cron.get_next(float)
 while True:
    now = time.time()
    if now >= next_run:
-        print(f"\n▶ [{time.strftime('%Y-%m-%d %H:%M:%S')}] Starting winnow...", flush=True)
+        print(f"\n[{time.strftime('%Y-%m-%d %H:%M:%S')}] Starting winnow run...", flush=True)
        check_models()
-        result = subprocess.run(["uv", "run", "winnow"], env=RUN_ENV)
+        try:
-        if result.returncode != 0:
+            main()
-            logger.error(f"winnow exited with code {result.returncode}")
+            print("winnow run complete", flush=True)
-            print(f"❌ winnow failed with exit code {result.returncode}", flush=True)
+        except KeyboardInterrupt:
-        else:
+            raise
-            print("✅ winnow completed successfully", flush=True)
+        except Exception as e:
            logger.error(f"winnow run failed: {e}", exc_info=True)
            print(f"winnow run failed: {e}", flush=True)
        next_run = cron.get_next(float)
-    time.sleep(60)
+    time.sleep(max(1, next_run - time.time()))
@@ -25,7 +25,7 @@ def test_config_loads_defaults(monkeypatch):
    assert cfg.FACE_MARGIN == 0.15
    assert cfg.USE_FULL_RESOLUTION is True
    assert cfg.ENABLE_FACE_ALIGNMENT is True
-    assert cfg.ENABLE_CACHE is False
+    assert cfg.ENABLE_CACHE is True
    _Config.reset()
@@ -0,0 +1,151 @@
 """Tests for image quality filtering functions."""
 import numpy as np
 from PIL import Image
 def _rgb_image(r, g, b, size=(100, 100)) -> Image.Image:
    arr = np.full((*size, 3), [r, g, b], dtype=np.uint8)
    return Image.fromarray(arr, "RGB")
 def _noisy_color_image(size=(100, 100)) -> Image.Image:
    """Noisy image with a strong red channel so grayscale check passes."""
    rng = np.random.default_rng(0)
    arr = rng.integers(0, 256, (*size, 3), dtype=np.uint8)
    arr[:, :, 0] = np.clip(arr[:, :, 0].astype(int) + 80, 0, 255).astype(np.uint8)
    arr[:, :, 2] = np.clip(arr[:, :, 2].astype(int) - 80, 0, 255).astype(np.uint8)
    return Image.fromarray(arr, "RGB")
 # ── check_blur ────────────────────────────────────────────────────────────────
 def test_blur_rejects_flat_image():
    from winnow.quality import check_blur
    flat = np.full((100, 100, 3), 128, dtype=np.uint8)
    passed, reason = check_blur(flat, threshold=100.0)
    assert not passed
    assert "Blurry" in reason
 def test_blur_passes_noisy_color_image():
    from winnow.quality import check_blur
    img = _noisy_color_image()
    passed, _ = check_blur(np.asarray(img), threshold=100.0)
    assert passed
 # ── check_grayscale ───────────────────────────────────────────────────────────
 def test_grayscale_rejects_ir_image():
    from winnow.quality import check_grayscale
    gray = np.full((100, 100, 3), 128, dtype=np.uint8)
    passed, reason = check_grayscale(gray)
    assert not passed
    assert "Grayscale" in reason
 def test_grayscale_passes_color_image():
    from winnow.quality import check_grayscale
    color = np.zeros((100, 100, 3), dtype=np.uint8)
    color[:, :, 0] = 200  # strong red channel
    passed, _ = check_grayscale(color)
    assert passed
 def test_grayscale_rejects_single_channel():
    from winnow.quality import check_grayscale
    single = np.full((100, 100, 1), 128, dtype=np.uint8)
    passed, reason = check_grayscale(single)
    assert not passed
 # ── check_exposure ────────────────────────────────────────────────────────────
 def test_exposure_rejects_black_image():
    from winnow.quality import check_exposure
    black = np.zeros((100, 100, 3), dtype=np.uint8)
    passed, reason = check_exposure(black)
    assert not passed
    assert "Underexposed" in reason
 def test_exposure_rejects_white_image():
    from winnow.quality import check_exposure
    white = np.full((100, 100, 3), 255, dtype=np.uint8)
    passed, reason = check_exposure(white)
    assert not passed
    assert "Overexposed" in reason
 def test_exposure_passes_normal_image():
    from winnow.quality import check_exposure
    mid = np.full((100, 100, 3), 128, dtype=np.uint8)
    passed, _ = check_exposure(mid)
    assert passed
 # ── check_face_size ───────────────────────────────────────────────────────────
 def test_face_size_rejects_small_face():
    from winnow.quality import check_face_size
    passed, reason = check_face_size(30, 30, min_px=50)
    assert not passed
    assert "small" in reason
 def test_face_size_passes_adequate_face():
    from winnow.quality import check_face_size
    passed, _ = check_face_size(100, 100, min_px=50)
    assert passed
 def test_face_size_rejects_if_either_dimension_small():
    from winnow.quality import check_face_size
    passed, _ = check_face_size(100, 30, min_px=50)
    assert not passed
 # ── check_confidence ──────────────────────────────────────────────────────────
 def test_confidence_rejects_low_score():
    from winnow.quality import check_confidence
    passed, reason = check_confidence(0.5, min_conf=0.7)
    assert not passed
    assert "confidence" in reason.lower()
 def test_confidence_passes_high_score():
    from winnow.quality import check_confidence
    passed, _ = check_confidence(0.95, min_conf=0.7)
    assert passed
 def test_confidence_passes_none_score():
    from winnow.quality import check_confidence
    passed, _ = check_confidence(None, min_conf=0.7)
    assert passed
 # ── assess_quality (integration) ─────────────────────────────────────────────
 def test_assess_quality_passes_good_image():
    from winnow.quality import assess_quality
    img = _noisy_color_image()
    result = assess_quality(img, face_bbox=(10, 10, 110, 110), confidence=0.9)
    assert result.passed
 def test_assess_quality_collects_multiple_failures():
    from winnow.quality import assess_quality
    black = _rgb_image(0, 0, 0)
    result = assess_quality(black, face_bbox=(0, 0, 10, 10), confidence=0.3)
    assert not result.passed
    assert len(result.reasons) >= 2
 def test_assess_quality_skips_face_size_without_bbox():
    from winnow.quality import assess_quality
    img = _noisy_color_image()
    result = assess_quality(img, face_bbox=None, confidence=0.9)
    assert result.passed
@@ -13,9 +13,13 @@ resolution-markers = [
    "platform_machine != 'aarch64' and platform_machine != 's390x' and platform_machine != 'x86_64' and sys_platform == 'linux'",
    "platform_machine == 's390x' and sys_platform == 'linux'",
 ]
 required-markers = [
    "platform_machine == 'x86_64' and sys_platform == 'linux'",
    "platform_machine == 'aarch64' and sys_platform == 'linux'",
 ]
 [manifest]
-overrides = [{ name = "onnxruntime-gpu", specifier = ">=1.23.2" }]
+overrides = [{ name = "onnxruntime", specifier = ">=1.23.2" }]
 [[package]]
 name = "annotated-doc"
@@ -1238,17 +1242,15 @@ name = "onnxruntime-gpu"
 version = "1.26.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "flatbuffers" },
+    { name = "flatbuffers", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "numpy" },
+    { name = "numpy", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "packaging" },
+    { name = "packaging", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "protobuf" },
+    { name = "protobuf", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
 ]
 wheels = [
    { url = "https://files.pythonhosted.org/packages/dd/97/fe8979f44b9275654b42f7bb556e30789b71a1b22998c83b540df2b1b774/onnxruntime_gpu-1.26.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:cfda2fad535595bfc3e570eb588092717711dcb2957656d814695e0c9ceb1508", size = 276974871, upload-time = "2026-05-08T19:15:58.052Z" },
    { url = "https://files.pythonhosted.org/packages/67/3f/59f1777a394625ecc9a85636de57dc47c25dbb5f888da050f1463955a0ce/onnxruntime_gpu-1.26.0-cp313-cp313-win_amd64.whl", hash = "sha256:6ab9f9c741d2e239b2e321ab0d389c04329d4ab7f11e3b92dd3aa7db1c59dee4", size = 226548083, upload-time = "2026-05-08T19:09:44.408Z" },
    { url = "https://files.pythonhosted.org/packages/89/96/360328e3c463f7ea08e853c4239c397e83363dd0204de71a710dc1a544bd/onnxruntime_gpu-1.26.0-cp313-cp313t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bcf6f347cad9f88a9a625c2b352cf9de927528aedb627ebbc089a201f1990b94", size = 276992052, upload-time = "2026-05-08T19:16:09.892Z" },
    { url = "https://files.pythonhosted.org/packages/fd/c8/aa2dc0e79bba577f37d5448bcb32fea79977e07506684d8138c19a0f1077/onnxruntime_gpu-1.26.0-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9e6e4fb1ec9ae1cf456534d9115f106ab2a1ae96fa513b4ed0f4795302b4a2c6", size = 276978254, upload-time = "2026-05-08T19:16:22.096Z" },
    { url = "https://files.pythonhosted.org/packages/41/e7/923298431e669567d7ccc2a4c898b6534a47641a051569fd97165fe6d9b8/onnxruntime_gpu-1.26.0-cp314-cp314-win_amd64.whl", hash = "sha256:3e592439b0183d303c2374517b5b392599a3d50b2dc9de949b9b15731ac921c9", size = 229142768, upload-time = "2026-05-08T19:09:54.589Z" },
    { url = "https://files.pythonhosted.org/packages/97/91/93ffe5431d154989f5e04864a25a97eea480997d771232bcbbc538188241/onnxruntime_gpu-1.26.0-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:56dc7b73954ff4bdc71f5b8ab306b6f61be5d007881b6ef423a609e2b9cd088b", size = 276991545, upload-time = "2026-05-08T19:16:33.347Z" },
 ]
@@ -1859,15 +1861,16 @@ version = "2.12.0"
 source = { registry = "https://download.pytorch.org/whl/cpu" }
 resolution-markers = [
    "platform_machine != 's390x' and sys_platform == 'darwin'",
    "platform_machine == 's390x' and sys_platform == 'darwin'",
 ]
 dependencies = [
-    { name = "filelock", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "filelock", marker = "sys_platform == 'darwin'" },
-    { name = "fsspec", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "fsspec", marker = "sys_platform == 'darwin'" },
-    { name = "jinja2", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "jinja2", marker = "sys_platform == 'darwin'" },
-    { name = "networkx", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "networkx", marker = "sys_platform == 'darwin'" },
-    { name = "setuptools", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "setuptools", marker = "sys_platform == 'darwin'" },
-    { name = "sympy", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "sympy", marker = "sys_platform == 'darwin'" },
-    { name = "typing-extensions", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "typing-extensions", marker = "sys_platform == 'darwin'" },
 ]
 wheels = [
    { url = "https://download-r2.pytorch.org/whl/cpu/torch-2.12.0-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:90dd587a5f61bfe1307148b581e2084fc5bc4a06e2b90a20e9a36b81087ff16b", upload-time = "2026-05-12T16:20:17Z" },
@@ -1922,16 +1925,15 @@ resolution-markers = [
    "platform_machine == 's390x' and sys_platform == 'win32'",
    "platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
    "platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
    "platform_machine == 's390x' and sys_platform == 'darwin'",
 ]
 dependencies = [
-    { name = "filelock", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "filelock", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "fsspec", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "fsspec", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "jinja2", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "jinja2", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "networkx", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "networkx", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "setuptools", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "setuptools", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "sympy", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "sympy", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "typing-extensions", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "typing-extensions", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
 ]
 wheels = [
    { url = "https://download-r2.pytorch.org/whl/cpu/torch-2.12.0%2Bcpu-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:68b7ddd4db4603a03e106e74c7098c8d8c8943d33c1e5ada009ca4cd885759c3", upload-time = "2026-05-12T23:17:12Z" },
@@ -1981,11 +1983,12 @@ version = "0.27.0"
 source = { registry = "https://download.pytorch.org/whl/cpu" }
 resolution-markers = [
    "platform_machine != 's390x' and sys_platform == 'darwin'",
    "platform_machine == 's390x' and sys_platform == 'darwin'",
 ]
 dependencies = [
-    { name = "numpy", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "numpy", marker = "sys_platform == 'darwin'" },
-    { name = "pillow", marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "pillow", marker = "sys_platform == 'darwin'" },
-    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform == 'darwin'" },
 ]
 wheels = [
    { url = "https://download-r2.pytorch.org/whl/cpu/torchvision-0.27.0-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:41d6dae73e1af09fa82ded597ae57f2a2314285acde54b25890a8f8e51b999d7", upload-time = "2026-05-12T16:20:37Z" },
@@ -2028,12 +2031,11 @@ resolution-markers = [
    "platform_machine == 's390x' and sys_platform == 'win32'",
    "platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
    "platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
    "platform_machine == 's390x' and sys_platform == 'darwin'",
 ]
 dependencies = [
-    { name = "numpy", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "numpy", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "pillow", marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "pillow", marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
-    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
 ]
 wheels = [
    { url = "https://download-r2.pytorch.org/whl/cpu/torchvision-0.27.0%2Bcpu-cp313-cp313-manylinux_2_28_aarch64.whl", hash = "sha256:69093b64b2762c43df17be2db2be163029963d90bc3f1801500fdeb723e54833", upload-time = "2026-05-12T16:20:36Z" },
@@ -2151,13 +2153,13 @@ dependencies = [
    { name = "pyyaml" },
    { name = "requests" },
    { name = "scipy" },
-    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform == 'darwin'" },
    { name = "torch", version = "2.12.0", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'aarch64' and platform_machine != 'x86_64' and sys_platform == 'linux'" },
-    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
    { name = "torch", version = "2.12.0+cu126", source = { registry = "https://download.pytorch.org/whl/cu126" }, marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "torchvision", version = "0.27.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "torchvision", version = "0.27.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform == 'darwin'" },
    { name = "torchvision", version = "0.27.0", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'aarch64' and platform_machine != 'x86_64' and sys_platform == 'linux'" },
-    { name = "torchvision", version = "0.27.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "torchvision", version = "0.27.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
    { name = "torchvision", version = "0.27.0+cu126", source = { registry = "https://download.pytorch.org/whl/cu126" }, marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
    { name = "ultralytics-thop" },
 ]
@@ -2172,9 +2174,9 @@ version = "2.0.20"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "numpy" },
-    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform == 'darwin'" },
    { name = "torch", version = "2.12.0", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'aarch64' and platform_machine != 'x86_64' and sys_platform == 'linux'" },
-    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
    { name = "torch", version = "2.12.0+cu126", source = { registry = "https://download.pytorch.org/whl/cu126" }, marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/98/c6/d25cb53e141242f74950744179c21b8798bb09b9e7161465ecda4f577ddf/ultralytics_thop-2.0.20.tar.gz", hash = "sha256:f3595e0d8c6fd0b9f62fc2cd9be921755e2649a05c34f1fabaea0bff7295d641", size = 34682, upload-time = "2026-06-06T11:42:42.184Z" }
@@ -2193,7 +2195,7 @@ wheels = [
 [[package]]
 name = "winnow"
-version = "0.2.8"
+version = "0.2.9"
 source = { editable = "." }
 dependencies = [
    { name = "croniter" },
@@ -2201,19 +2203,20 @@ dependencies = [
    { name = "numpy" },
    { name = "nvidia-cudnn-cu12", version = "9.10.2.21", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
    { name = "nvidia-cudnn-cu12", version = "9.23.1.3", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'x86_64' or sys_platform != 'linux'" },
-    { name = "onnxruntime-gpu" },
+    { name = "onnxruntime" },
    { name = "onnxruntime-gpu", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
    { name = "opencv-python-headless" },
    { name = "pillow" },
    { name = "python-dotenv" },
    { name = "requests" },
    { name = "rich" },
-    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "torch", version = "2.12.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform == 'darwin'" },
    { name = "torch", version = "2.12.0", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'aarch64' and platform_machine != 'x86_64' and sys_platform == 'linux'" },
-    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "torch", version = "2.12.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
    { name = "torch", version = "2.12.0+cu126", source = { registry = "https://download.pytorch.org/whl/cu126" }, marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
-    { name = "torchvision", version = "0.27.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "platform_machine != 's390x' and sys_platform == 'darwin'" },
+    { name = "torchvision", version = "0.27.0", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "sys_platform == 'darwin'" },
    { name = "torchvision", version = "0.27.0", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'aarch64' and platform_machine != 'x86_64' and sys_platform == 'linux'" },
-    { name = "torchvision", version = "0.27.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 's390x' and sys_platform == 'darwin') or (platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
+    { name = "torchvision", version = "0.27.0+cpu", source = { registry = "https://download.pytorch.org/whl/cpu" }, marker = "(platform_machine == 'aarch64' and sys_platform == 'linux') or (sys_platform != 'darwin' and sys_platform != 'linux')" },
    { name = "torchvision", version = "0.27.0+cu126", source = { registry = "https://download.pytorch.org/whl/cu126" }, marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
    { name = "transformers" },
    { name = "ultralytics" },
@@ -2231,7 +2234,9 @@ requires-dist = [
    { name = "insightface", specifier = ">=0.7.3" },
    { name = "numpy", specifier = ">=2.2.6" },
    { name = "nvidia-cudnn-cu12", specifier = ">=9.0.0" },
-    { name = "onnxruntime-gpu", specifier = ">=1.23.2" },
+    { name = "onnxruntime", marker = "sys_platform != 'linux'", specifier = ">=1.23.2" },
    { name = "onnxruntime", marker = "platform_machine != 'x86_64' and sys_platform == 'linux'", specifier = ">=1.23.2" },
    { name = "onnxruntime-gpu", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'", specifier = ">=1.23.2" },
    { name = "opencv-python-headless", specifier = ">=4.12.0.88" },
    { name = "pillow", specifier = ">=12.1.0" },
    { name = "python-dotenv", specifier = ">=1.2.1" },
@@ -5,4 +5,9 @@ Immich library for Frigate's Face Recognition (ArcFace) and Object/State
 Classification models.
 """
-__version__ = "0.1.0"
+from importlib.metadata import PackageNotFoundError, version
 try:
    __version__ = version("winnow")
 except PackageNotFoundError:
    __version__ = "unknown"
@@ -2,6 +2,7 @@
 import logging
 import os
 import sys
 from rich import print as rprint
 from rich.prompt import Confirm
@@ -10,7 +11,7 @@ from .config import Config, ConfigManager
 from .executor import execute_jobs, upload_to_frigate
 from .immich_api import get_people
 from .jobs import _show_preview, auto_configure, interactive_configure
-from .logging import console, setup_logging
+from .log_config import console, setup_logging
 from .upload_tracker import get_person_summary, reset_person
 logger = logging.getLogger(__name__)
@@ -19,7 +20,8 @@ logger = logging.getLogger(__name__)
 def main() -> None:
    """Entry point for winnow CLI."""
    try:
-        setup_logging(verbose=False)
+        verbose = os.environ.get("VERBOSE", "").lower() in ("true", "1", "yes")
        setup_logging(verbose=verbose)
        console.print(r"""
    [bold blue]winnow[/bold blue]
@@ -63,17 +65,18 @@ def main() -> None:
            rprint("[bold red]Could not fetch people from Immich. Check URL/Key.[/bold red]")
            return
-        # Check for non-interactive mode
+        # Auto mode when no TTY (Docker, cron, pipes) — the primary use case.
-        auto_mode = os.environ.get("AUTO_MODE", "false").lower() == "true"
+        # A TTY means local interactive use; AUTO_MODE=true overrides that for scripting.
        auto_mode = not sys.stdin.isatty() or os.environ.get("AUTO_MODE", "").lower() in ("true", "1", "yes")
        dry_run = os.environ.get("DRY_RUN", "false").lower() in ("true", "1", "yes")
        if dry_run:
            rprint("[bold yellow]DRY RUN — no images will be downloaded or uploaded[/bold yellow]")
        if auto_mode:
            rprint("[bold cyan]Running in AUTO mode (non-interactive)[/bold cyan]")
            jobs = auto_configure(people)
        else:
            rprint("[bold cyan]Interactive mode — set AUTO_MODE=true to skip prompts[/bold cyan]")
            jobs = interactive_configure(people)
        if jobs:
@@ -39,8 +39,7 @@ class _Config:
    USE_FULL_RESOLUTION: bool = True
    ENABLE_FACE_ALIGNMENT: bool = True
-    # Caching (opt-in to avoid unexpected files)
+    ENABLE_CACHE: bool = True
    ENABLE_CACHE: bool = False
    CACHE_DIR: str = ".if_cache"
    def __new__(cls) -> "_Config":
@@ -64,7 +63,7 @@ class _Config:
        self.FACE_MARGIN = float(os.getenv("FACE_MARGIN", "0.15"))
        self.USE_FULL_RESOLUTION = os.getenv("USE_FULL_RESOLUTION", "true").lower() in ("true", "1", "yes")
        self.ENABLE_FACE_ALIGNMENT = os.getenv("ENABLE_FACE_ALIGNMENT", "true").lower() in ("true", "1", "yes")
-        self.ENABLE_CACHE = os.getenv("ENABLE_CACHE", "false").lower() in ("true", "1", "yes")
+        self.ENABLE_CACHE = os.getenv("ENABLE_CACHE", "true").lower() in ("true", "1", "yes")
        self.CACHE_DIR = os.getenv("CACHE_DIR", ".if_cache")
        # Fall back to config file for non-sensitive values (API_KEY not stored here)
@@ -161,7 +160,12 @@ class _ConfigAccessor:
 Config = _ConfigAccessor()
-ConfigManager = type("ConfigManager", (), {"get": staticmethod(lambda: _Config())})
+
 class ConfigManager:
    @staticmethod
    def get() -> _Config:
        return _Config()
 def get_headers() -> dict[str, str]:
@@ -29,6 +29,7 @@ def select_diverse_assets(
    entity_name: str,
    selection_mode: str = "smart",
    entity_type: str = "face",
    person_id: str | None = None,
    progress_callback=None,
 ) -> list:
    """
@@ -59,7 +60,7 @@ def select_diverse_assets(
        return _select_time_spread(assets, limit)
    try:
-        return _select_by_embedding(assets, limit, entity_type, progress_callback)
+        return _select_by_embedding(assets, limit, entity_type, person_id, progress_callback)
    except Exception as e:
        logger.error(f"Smart Diversity failed: {e}. Falling back to time spread.")
        return _select_time_spread(assets, limit)
@@ -80,9 +81,11 @@ def _fetch_thumbnail(asset_id: str, timeout: int = 10) -> Image.Image | None:
        return None
-def _get_face_bbox(asset: dict) -> tuple[float, float, float, float] | None:
+def _get_face_bbox(asset: dict, person_id: str | None = None) -> tuple[float, float, float, float] | None:
-    """Extract face bounding box from asset metadata if available."""
+    """Extract face bounding box from asset metadata for the given person."""
    for person in asset.get("people", []):
        if person_id and person.get("id") != person_id:
            continue
        faces = person.get("faces", [])
        if faces:
            f = faces[0]
@@ -95,12 +98,16 @@ def _get_face_bbox(asset: dict) -> tuple[float, float, float, float] | None:
    return None
-def _get_face_confidence(asset: dict) -> float | None:
+def _get_face_confidence(asset: dict, person_id: str | None = None) -> float | None:
-    """Extract face detection confidence from asset metadata if available."""
+    """Extract face detection confidence from asset metadata for the given person."""
    for person in asset.get("people", []):
        if person_id and person.get("id") != person_id:
            continue
        faces = person.get("faces", [])
        if faces:
-            return faces[0].get("score") or faces[0].get("confidence")
+            f = faces[0]
            score = f.get("score")
            return score if score is not None else f.get("confidence")
    return None
@@ -108,6 +115,7 @@ def _crop_face_from_thumbnail(
    img: Image.Image,
    asset: dict,
    margin: float = 0.25,
    person_id: str | None = None,
 ) -> Image.Image | None:
    """Crop the face region from a thumbnail using Immich bbox metadata.
@@ -118,19 +126,22 @@ def _crop_face_from_thumbnail(
        img: Full preview thumbnail
        asset: Asset dict with people/faces metadata
        margin: Extra margin around the bbox (fraction, default 25%)
        person_id: If provided, only crop from this person's face data.
    Returns:
        Cropped face PIL image, or None if no face metadata available
    """
-    bbox = _get_face_bbox(asset)
+    bbox = _get_face_bbox(asset, person_id=person_id)
    if bbox is None:
        return None
    x1, y1, x2, y2 = bbox
    img_w, img_h = img.size
-    # Get metadata dimensions to scale bbox
+    # Get metadata dimensions to scale bbox — must match the same person as _get_face_bbox
    for person in asset.get("people", []):
        if person_id and person.get("id") != person_id:
            continue
        faces = person.get("faces", [])
        if faces:
            meta_w = faces[0].get("imageWidth") or img_w
@@ -169,6 +180,7 @@ def _select_by_embedding(
    assets: list,
    limit: int | str,
    entity_type: str,
    person_id: str | None = None,
    progress_callback=None,
 ) -> list:
    """Select assets using embedding-based cluster-aware FPS.
@@ -217,11 +229,11 @@ def _select_by_embedding(
        if img is None:
            continue
-        confidence = _get_face_confidence(asset)
+        confidence = _get_face_confidence(asset, person_id=person_id)
        # Quality gate: filter before expensive embedding computation
        if entity_type == "face":
-            face_bbox = _get_face_bbox(asset)
+            face_bbox = _get_face_bbox(asset, person_id=person_id)
            quality = assess_quality(
                img,
                face_bbox=face_bbox,
@@ -236,7 +248,7 @@ def _select_by_embedding(
                continue
            # Crop the target person's face before embedding
-            face_crop = _crop_face_from_thumbnail(img, asset)
+            face_crop = _crop_face_from_thumbnail(img, asset, person_id=person_id)
            embed_img = face_crop if face_crop is not None else img
        else:
            embed_img = img
@@ -360,7 +372,7 @@ def _compute_adaptive_threshold(emb_normed: np.ndarray, entity_type: str) -> flo
    fraction = 0.20 if entity_type == "face" else 0.10
    threshold = max(0.05, median_dist * fraction)
-    logger.info(
+    logger.debug(
        f"Adaptive threshold: {threshold:.4f} "
        f"(median_dist={median_dist:.4f}, fraction={fraction}, type={entity_type})"
    )
@@ -402,7 +414,7 @@ def _cluster_aware_selection(
    # --- Stage 1: K-Medoids clustering ---
    k = min(max(5, target // 4), n // 3, n)  # e.g., 5-20 clusters
-    logger.info(f"Clustering {n} embeddings into {k} groups (K-Medoids)...")
+    logger.debug(f"Clustering {n} embeddings into {k} groups (K-Medoids)...")
    # Compute full cosine distance matrix
    dist_matrix = 1 - emb_normed @ emb_normed.T
@@ -411,7 +423,7 @@ def _cluster_aware_selection(
    selected = list(medoid_indices)
    selected_set = set(selected)
-    logger.info(f"Selected {len(selected)} cluster medoids as initial picks.")
+    logger.debug(f"Selected {len(selected)} cluster medoids as initial picks.")
    # --- Stage 2: FPS with hard example weighting ---
    min_dists = np.full(n, np.inf)
@@ -436,8 +448,8 @@ def _cluster_aware_selection(
            break  # All points selected
        if limit == "auto" and best_dist < auto_threshold:
-            logger.info(
+            logger.debug(
-                f"Auto-stop: Next best image {best_dist:.3f} away " f"(adaptive threshold {auto_threshold:.4f})."
+                f"Auto-stop: next best image {best_dist:.3f} away (adaptive threshold {auto_threshold:.4f})."
            )
            break
@@ -6,11 +6,13 @@ Unified embedding interface for faces and objects.
 - Caching: Disk-based cache avoids recomputation on reruns
 """
 import contextlib
 import importlib
 import logging
 import os
 import time
 import warnings
 from contextlib import contextmanager
 from pathlib import Path
 import cv2
 import numpy as np
@@ -20,6 +22,26 @@ from .cache import get_cache
 logger = logging.getLogger(__name__)
@contextmanager
 def _suppress_output():
    """Suppress stdout/stderr at the file-descriptor level, silencing C extension noise."""
    devnull_fd = os.open(os.devnull, os.O_WRONLY)
    saved_out, saved_err = os.dup(1), os.dup(2)
    try:
        os.dup2(devnull_fd, 1)
        os.dup2(devnull_fd, 2)
        yield
    finally:
        try:
            os.dup2(saved_out, 1)
        finally:
            os.dup2(saved_err, 2)
        os.close(devnull_fd)
        os.close(saved_out)
        os.close(saved_err)
 # Lazy-loaded singletons
 _insightface_app = None
 _insightface_loaded = False
@@ -37,18 +59,14 @@ def _preload_cuda_libs() -> None:
    """Preload CUDA/cuDNN DLLs so onnxruntime-gpu registers CUDAExecutionProvider.
    Starting with onnxruntime-gpu 1.19+, CUDA/cuDNN libraries are no longer
-    bundled inside the ORT package. They must be loaded from the nvidia-*
+    bundled inside the ORT package — they come from the nvidia-* pip packages.
-    pip packages (nvidia-cuda-runtime-cu12, nvidia-cudnn-cu12) before any
+    preload_dlls() locates them automatically via site-packages discovery.
    InferenceSession is created.
    Calling preload_dlls() with directory="" searches NVIDIA site-packages
    directories automatically.
    """
    try:
        import onnxruntime
        if hasattr(onnxruntime, "preload_dlls"):
-            onnxruntime.preload_dlls(cuda=True, cudnn=True, directory="")
+            onnxruntime.preload_dlls(cuda=True, cudnn=True)
-            logger.info("Preloaded CUDA/cuDNN DLLs for onnxruntime-gpu")
+            logger.debug("Preloaded CUDA/cuDNN DLLs for onnxruntime-gpu")
        else:
            logger.debug("onnxruntime.preload_dlls() not available (ORT < 1.21)")
    except Exception as e:
@@ -70,32 +88,48 @@ def get_insightface_app():
    # Preload CUDA/cuDNN DLLs BEFORE any ORT InferenceSession is created
    _preload_cuda_libs()
    ctx_id = -1
    insightface_home = os.environ.get("INSIGHTFACE_HOME", os.path.expanduser("~/.insightface"))
    try:
        import onnxruntime as ort
        from insightface.app import FaceAnalysis
        # Disk cache check — lets the user know whether a download is coming
        buffalo_path = Path(insightface_home) / "models" / "buffalo_l"
        if buffalo_path.exists() and any(buffalo_path.iterdir()):
            logger.info("InsightFace Buffalo_L: found in model cache")
        else:
            logger.info("InsightFace Buffalo_L: not cached — downloading now (~300 MB)")
        # Get providers, excluding TensorRT to avoid noisy errors
        providers = [p for p in ort.get_available_providers() if p != "TensorrtExecutionProvider"]
-        logger.info(f"Available ONNX providers: {providers}")
+        logger.debug(f"ONNX providers available: {providers}")
        # Determine device: 0 for GPU, -1 for CPU
        gpu_providers = {
            "CUDAExecutionProvider",
            "ROCmExecutionProvider",
            "MPSExecutionProvider",
            "CoreMLExecutionProvider",
        }
-        ctx_id = -1 if _is_force_cpu() else (0 if gpu_providers & set(providers) else -1)
+        has_gpu_provider = bool(gpu_providers & set(providers))
        ctx_id = -1 if _is_force_cpu() else (0 if has_gpu_provider else -1)
        if not has_gpu_provider and not _is_force_cpu():
            logger.warning(
                "No GPU execution provider found — running InsightFace on CPU. "
                "If you have an NVIDIA GPU, ensure the NVIDIA Container Toolkit is "
                "installed and the container has GPU access (deploy.resources in compose)."
            )
        device_str = "GPU" if ctx_id >= 0 else "CPU"
-        logger.info(f"Loading InsightFace Buffalo_L on {device_str} (ctx_id={ctx_id})...")
+        logger.info(f"InsightFace Buffalo_L: loading into memory on {device_str}...")
-        # Suppress C-level output during model loading
+        t0 = time.time()
-        with open(os.devnull, "w") as devnull, contextlib.redirect_stdout(devnull), contextlib.redirect_stderr(devnull):
+        with _suppress_output():
            insightface_home = os.environ.get("INSIGHTFACE_HOME", os.path.expanduser("~/.insightface"))
            _insightface_app = FaceAnalysis(name="buffalo_l", root=insightface_home, providers=providers)
            _insightface_app.prepare(ctx_id=ctx_id, det_size=(640, 640))
        logger.info(f"InsightFace Buffalo_L: ready on {device_str} ({time.time() - t0:.1f}s)")
        return _insightface_app
    except ImportError:
@@ -103,17 +137,23 @@ def get_insightface_app():
        return None
    except Exception as e:
        logger.error(f"Failed to load InsightFace: {e}")
        # Retry on CPU if GPU failed
        if ctx_id == 0:
-            logger.warning("Retrying InsightFace on CPU...")
+            logger.warning("InsightFace GPU load failed — retrying on CPU...")
            try:
                from insightface.app import FaceAnalysis
-                _insightface_app = FaceAnalysis(name="buffalo_l", root=insightface_home)
+                t0 = time.time()
-                _insightface_app.prepare(ctx_id=-1, det_size=(640, 640))
+                with _suppress_output():
                    _insightface_app = FaceAnalysis(
                        name="buffalo_l",
                        root=insightface_home,
                        providers=["CPUExecutionProvider"],
                    )
                    _insightface_app.prepare(ctx_id=-1, det_size=(640, 640))
                logger.info(f"InsightFace Buffalo_L: ready on CPU (fallback, {time.time() - t0:.1f}s)")
                return _insightface_app
            except Exception as ex:
-                logger.error(f"CPU fallback failed: {ex}")
+                logger.error(f"InsightFace CPU fallback failed: {ex}")
        return None
@@ -162,7 +202,18 @@ def get_siglip_model():
        from transformers import AutoImageProcessor, SiglipVisionModel
        model_name = "google/siglip-base-patch16-224"
-        logger.info(f"Loading SigLIP model ({model_name})...")
+
        # Disk cache check — path derived from model_name using HuggingFace's slug convention
        hf_home = os.environ.get("HF_HOME", os.path.join(os.path.expanduser("~"), ".cache", "huggingface"))
        cache_slug = "models--" + model_name.replace("/", "--")
        model_cache = Path(hf_home) / "hub" / cache_slug
        if model_cache.exists() and any(model_cache.iterdir()):
            logger.info(f"SigLIP {model_name}: found in model cache")
        else:
            logger.info(f"SigLIP {model_name}: not cached — downloading now (~380 MB)")
        logger.info(f"SigLIP {model_name}: loading into memory...")
        t0 = time.time()
        with warnings.catch_warnings():
            warnings.filterwarnings("ignore", category=FutureWarning)
@@ -176,15 +227,16 @@ def get_siglip_model():
        if not _is_force_cpu():
            if torch.cuda.is_available():
                _siglip_model = _siglip_model.cuda()
-                logger.info("SigLIP running on CUDA GPU")
+                device_name = "CUDA GPU"
            elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
                _siglip_model = _siglip_model.to("mps")
-                logger.info("SigLIP running on Apple MPS")
+                device_name = "Apple MPS"
            else:
-                logger.info("SigLIP running on CPU")
+                device_name = "CPU"
        else:
-            logger.info("FORCE_CPU set. SigLIP running on CPU")
+            device_name = "CPU (FORCE_CPU)"
        logger.info(f"SigLIP {model_name}: ready on {device_name} ({time.time() - t0:.1f}s)")
        return _siglip_model, _siglip_processor
    except ImportError as e:
@@ -267,30 +319,31 @@ def get_embedding(
    use_cache = Config.ENABLE_CACHE and asset_id is not None
    cache = get_cache(Config.CACHE_DIR) if use_cache else None
-    model_key = "immich" if entity_type == "face" else "siglip"
+    # Use a single consistent cache key per model so lookups and stores always match.
    # "immich" was previously used as the face key on the lookup path but "insightface"
    # on the store path — meaning the cache was never hit for locally-computed embeddings.
    cache_key = "insightface" if entity_type == "face" else "siglip"
    # 1. Use Immich embedding if provided
    if immich_embedding is not None:
        if cache:
-            cache.put(asset_id, immich_embedding, model_key)
+            cache.put(asset_id, immich_embedding, cache_key)
        return immich_embedding
    # 2. Check disk cache
    if cache:
-        cached = cache.get(asset_id, model_key)
+        cached = cache.get(asset_id, cache_key)
        if cached is not None:
            return cached
    # 3. Compute locally
    if entity_type == "face":
        emb = get_face_embedding(img_pil)
        model_key = "insightface"
    else:
        emb = get_object_embedding(img_pil)
    # Cache the result
    if emb is not None and cache:
-        cache.put(asset_id, emb, model_key)
+        cache.put(asset_id, emb, cache_key)
    return emb
@@ -14,7 +14,7 @@ from rich.progress import BarColumn, Progress, SpinnerColumn, TaskProgressColumn
 from .config import Config, get_headers
 from .image_processing import process_face_mode, process_full_mode, process_object_mode
 from .immich_api import fetch_face_data, fetch_full_image
-from .logging import console
+from .log_config import console
 from .upload_tracker import mark_rejected, mark_uploaded
 logger = logging.getLogger(__name__)
@@ -21,8 +21,13 @@ def get_frigate_face_counts() -> dict[str, int] | None:
        resp = requests.get(f"{frigate_url}/api/faces", timeout=10)
        resp.raise_for_status()
        data = resp.json()
-        train = data.get("train", {})
+        # Response: {person_name: [file, ...], "train": [...], ...}
-        return {name: len(files) for name, files in train.items() if isinstance(files, list)}
+        # "train" is a flat pending list, not a person — skip it.
        return {
            name: len(files)
            for name, files in data.items()
            if name != "train" and isinstance(files, list)
        }
    except Exception as e:
        logger.warning(f"Could not query Frigate face counts: {e}")
        return None
@@ -35,10 +35,13 @@ def get_people() -> list[dict]:
            headers=get_headers(),
            timeout=10,
        )
        if resp.status_code == 401:
            logger.error("Immich API key is invalid or expired (401 Unauthorized). Update API_KEY.")
            return []
        resp.raise_for_status()
        return resp.json().get("people", [])
    except (requests.RequestException, ValueError) as e:
-        logger.error(f"Failed to fetch people: {e}")
+        logger.error(f"Failed to fetch people from Immich: {e}")
        return []
@@ -49,7 +52,7 @@ def fetch_all_assets(person: dict) -> list[dict]:
    url = f"{Config.IMMICH_URL}/api/search/metadata"
    page_size = 1000
-    logger.info(f"Fetching assets for {name}...")
+    logger.debug(f"Fetching assets for {name}...")
    assets = []
    for page in range(1, MAX_PAGES + 1):
@@ -137,10 +140,11 @@ def fetch_face_data(asset_id: str, person_id: str | None = None) -> FaceData | N
            face.get("boundingBoxY2", 0),
        )
        score = face.get("score")
        return FaceData(
            embedding=embedding,
            bbox=bbox,
-            confidence=face.get("score") or face.get("confidence"),
+            confidence=score if score is not None else face.get("confidence"),
            image_width=face.get("imageWidth", 0),
            image_height=face.get("imageHeight", 0),
        )
@@ -212,6 +216,6 @@ def filter_recent_assets(assets: list[dict], years: int | None = None) -> list[d
        except ValueError:
            continue
-    logger.info(f"Retained {len(recent)} assets (filtered {skipped} old assets).")
+    logger.debug(f"Retained {len(recent)} assets (filtered {skipped} old assets).")
    return recent
@@ -13,7 +13,7 @@ from .diversity import select_diverse_assets
 from .embeddings import is_embedding_available, load_embedding_model
 from .frigate_api import get_frigate_face_counts
 from .immich_api import fetch_all_assets, filter_recent_assets
-from .logging import console
+from .log_config import console
 from .upload_tracker import filter_already_uploaded, get_person_summary, update_frigate_count
 logger = logging.getLogger(__name__)
@@ -81,7 +81,9 @@ def _resolve_strategy(strategy: str, has_embedding: bool) -> tuple[int | str, st
    return strategy_map.get(strategy, ("auto", "smart"))
-def _perform_selection(assets: list, limit: int | str, name: str, selection_mode: str, entity_type: str) -> list:
+def _perform_selection(
    assets: list, limit: int | str, name: str, selection_mode: str, entity_type: str, person_id: str | None = None
 ) -> list:
    """Run diversity selection with progress display."""
    if selection_mode == "smart":
        model_display = "InsightFace (face embeddings)" if entity_type == "face" else "SigLIP (visual embeddings)"
@@ -104,6 +106,7 @@ def _perform_selection(assets: list, limit: int | str, name: str, selection_mode
                name,
                selection_mode=selection_mode,
                entity_type=entity_type,
                person_id=person_id,
                progress_callback=lambda c, t: progress.update(task, completed=c, total=t),
            )
@@ -113,7 +116,9 @@ def _perform_selection(assets: list, limit: int | str, name: str, selection_mode
    rprint(f"\n[cyan]Using time-spread selection for {limit} images...[/cyan]")
    with console.status(f"[bold]Selecting {limit} images evenly distributed over time...[/bold]"):
-        selected = select_diverse_assets(assets, limit, name, selection_mode="time", entity_type=entity_type)
+        selected = select_diverse_assets(
            assets, limit, name, selection_mode="time", entity_type=entity_type, person_id=person_id
        )
    rprint(f"  [green]Selected {len(selected)} images using time spread.[/green]")
    return selected
@@ -145,8 +150,11 @@ def _configure_person(person: dict, people: list[dict]) -> dict | None:
    rprint(f"  Found [bold]{len(all_assets)}[/bold] total, [bold]{len(recent_assets)}[/bold] in range ({years} years).")
-    # Filter out assets already uploaded to Frigate
+    # Filter out assets already uploaded to Frigate.
-    retry_rejected = os.environ.get("RETRY_REJECTED", "false").lower() in ("true", "1", "yes")
+    # In interactive mode, ask — use the env var only as the default so it can
    # still be pre-set (e.g. RETRY_REJECTED=true) without forcing the answer.
    retry_env = os.environ.get("RETRY_REJECTED", "false").lower() in ("true", "1", "yes")
    retry_rejected = Confirm.ask("Include previously rejected images?", default=retry_env)
    before_dedup = len(recent_assets)
    new_asset_ids = set(filter_already_uploaded([a["id"] for a in recent_assets], retry_rejected=retry_rejected))
    recent_assets = [a for a in recent_assets if a["id"] in new_asset_ids]
@@ -167,7 +175,9 @@ def _configure_person(person: dict, people: list[dict]) -> dict | None:
        return None
    # Perform selection
-    selected_assets = _perform_selection(recent_assets, limit, name, selection_mode, entity_type)
+    selected_assets = _perform_selection(
        recent_assets, limit, name, selection_mode, entity_type, person_id=person["id"]
    )
    rprint(f"  [green]Queued {len(selected_assets)} images for {name}.[/green]")
    return {"person": person, "assets": selected_assets, "limit": len(selected_assets), "config": config}
@@ -276,10 +286,8 @@ def auto_configure(people: list[dict]) -> list[dict]:
        if frigate_counts is not None:
            already_uploaded = frigate_counts.get(name, 0)
        else:
-            already_uploaded = (
+            fc = person_summary.get("frigate_count")
-                person_summary.get("frigate_count")
+            already_uploaded = fc if fc is not None else person_summary.get("uploaded", 0)
                or person_summary.get("uploaded", 0)
            )
        capacity = Config.MAX_AUTO_IMAGES - already_uploaded
        if capacity <= 0:
            rprint(
@@ -291,17 +299,24 @@ def auto_configure(people: list[dict]) -> list[dict]:
        has_embedding = is_embedding_available(entity_type)
        limit, selection_mode = _resolve_strategy(strategy, has_embedding)
-        # Cap selection to remaining capacity
+        # Cap selection to remaining capacity.
        # For auto mode with partial training, keep "auto" so adaptive stopping
        # still runs — just trim the result to the remaining capacity afterward.
        auto_cap = None
        if limit == "auto":
            if already_uploaded > 0:
-                limit = capacity  # partially filled — select exactly what remains
+                auto_cap = capacity
        else:
            limit = min(limit, capacity)
        if selection_mode == "skip":
            continue
-        selected_assets = _perform_selection(recent_assets, limit, name, selection_mode, entity_type)
+        selected_assets = _perform_selection(
            recent_assets, limit, name, selection_mode, entity_type, person_id=person["id"]
        )
        if auto_cap is not None:
            selected_assets = selected_assets[:auto_cap]
        if selected_assets:
            rprint(f"  [green]Queued {len(selected_assets)} images for {name}.[/green]")
@@ -26,10 +26,12 @@ def setup_logging(verbose: bool = False) -> logging.Logger:
    """Configure logging with Rich console and file output."""
    level = logging.DEBUG if verbose else logging.INFO
-    # Configure root logger
+    # Configure root logger; close existing handlers before replacing them
    root = logging.getLogger()
    root.setLevel(level)
-    root.handlers.clear()
+    for h in root.handlers[:]:
        h.close()
        root.removeHandler(h)
    # Rich console handler - uses shared console to avoid breaking progress bars
    root.addHandler(RichHandler(rich_tracebacks=True, markup=True, console=console))
@@ -37,7 +39,7 @@ def setup_logging(verbose: bool = False) -> logging.Logger:
    # File handler (always debug level) — log file respects OUTPUT_DIR if set
    log_dir = os.environ.get("OUTPUT_DIR", ".")
    os.makedirs(log_dir, exist_ok=True)
-    log_path = os.path.join(log_dir, "immich_export.log")
+    log_path = os.path.join(log_dir, "winnow.log")
    file_handler = logging.FileHandler(log_path)
    file_handler.setLevel(logging.DEBUG)
    file_handler.setFormatter(logging.Formatter("%(asctime)s - %(name)s - %(levelname)s - %(message)s"))
Author	SHA1	Message	Date
flanandClaude Sonnet 4.6	32e4235384	docs: add 0.2.10 changelog Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 16:51:53 +00:00
flanandClaude Sonnet 4.6	85e499a677	fix: audit — score falsy-zero, crop person mismatch, adaptive cap, suppress_output, ldconfig glob, scheduler sleep - immich_api: `score or confidence` treated 0.0 score as falsy; use explicit None check - diversity: same falsy-zero fix in _get_face_confidence - diversity: _crop_face_from_thumbnail scale loop now filters by person_id (was using first person's imageWidth/imageHeight regardless of target in group photos) - jobs: partially-trained auto mode kept limit="auto" for adaptive stopping, then caps result to remaining capacity (was converting to int, silently disabling FPS adaptive threshold and early-stop) - embeddings: _suppress_output finally block wraps first dup2 in try/finally so stderr is always restored even if stdout restore raises OSError - Dockerfile: ldconfig find uses python3.* glob instead of hardcoded python3.13 - scheduler: sleep until next_run instead of fixed 60s; eliminates late-fire jitter and unnecessary wakeups on long schedules Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 16:48:24 +00:00
flanandClaude Sonnet 4.6	ad4d212df0	chore: bump version to 0.2.10 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 16:38:40 +00:00
flanandClaude Sonnet 4.6	31e7ca5af0	fix: Frigate face count API structure and Immich 401 error message Frigate /api/faces response has person names as top-level keys with lists of filenames — {person: [file, ...], "train": [...]}. The old code incorrectly looked inside data["train"] as if it were a dict of persons, causing 'list object has no attribute items' on every run. Immich get_people() now checks for 401 before raise_for_status() and logs a clear "API key invalid or expired" message instead of the raw requests exception string. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:58:29 +00:00
flanandClaude Sonnet 4.6	4f6967258f	docs: update .env.example and README to reflect VERBOSE and current defaults - .env.example: add VERBOSE, remove active AUTO_MODE=true (now TTY-detected), comment out FORCE_CPU/ENABLE_CACHE default values, update CRON_SCHEDULE section to document all three modes - README: clarify that log file is always DEBUG regardless of VERBOSE Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:50:51 +00:00
flanandClaude Sonnet 4.6	882af37d8e	feat: VERBOSE=true enables DEBUG-level console output The log file already captures DEBUG unconditionally. This env var wires the same to the Rich console handler for troubleshooting without needing to read the log file. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:49:28 +00:00
flanandClaude Sonnet 4.6	cad053e88f	fix: add PYTHONPATH=/app so winnow entry point finds its package uv sync runs before winnow/ is COPY'd into the build stage, so the wheel uv builds contains only dist-info (no Python files). The console_scripts entry point sets sys.path[0] to its own directory (/app/.venv/bin), not /app, so 'from winnow.cli import main' fails at container startup. PYTHONPATH=/app makes the package importable regardless of how Python is invoked (script, -m, entry point, docker exec). This is preferable to re-ordering the COPY layers, which would bust the heavy uv sync cache on every winnow/ source change. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:44:56 +00:00
flanandClaude Sonnet 4.6	3db52c2141	fix+logging: model load logging, CPU fallback, log level audit Bugs fixed (from high-effort review): - InsightFace CPU fallback now passes providers=['CPUExecutionProvider'] and wraps in _suppress_output() so broken GPU drivers don't cause fallback to try the same broken provider again, and C-extension noise stays suppressed - scheduler.py: BaseException → Exception (KeyboardInterrupt already re-raised; winnow has no sys.exit() calls, so SystemExit would not occur, but Exception is the correct scope) - compose.yml: fix inverted AUTO_MODE comment (docker run -it enables interactive mode via TTY, not non-interactive) Model loading logging (embeddings.py): - InsightFace: disk cache check, "not cached — downloading now (~300 MB)", "loading into memory on GPU/CPU...", "ready on GPU/CPU (Xs)" - SigLIP: same treatment; cache path derived dynamically from model_name via HuggingFace slug convention (models--org--model) so it stays correct if the model variant ever changes Logging level audit (INFO/DEBUG/WARNING/ERROR): - diversity.py: internal algo steps (clustering, medoids, adaptive threshold, auto-stop decision) → DEBUG; final selection summaries stay INFO - immich_api.py: "Fetching assets" and "Retained N assets" → DEBUG (callers already print this to the console via rprint) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:23:44 +00:00
flanandClaude Sonnet 4.6	af8bae1c45	fix: embedding cache key mismatch, scheduler path/exception, log handler leak embeddings.py: face embedding cache used 'immich' for lookup but 'insightface' for storage, so the cache was never hit for locally- computed embeddings. Unified to 'insightface'/'siglip' throughout. This affects all users since ENABLE_CACHE now defaults to true. scheduler.py: INSIGHTFACE_HOME=/models/.insightface was having '.insightface' appended again, making buffalo_l check always report 'will download'. Also catch BaseException (not just Exception) so a SystemExit from a library call can't silently kill all future runs. log_config.py: handlers.clear() abandoned open FileHandler fds on each scheduled main() call. Close each handler properly before removal. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:10:44 +00:00
flanandClaude Sonnet 4.6	ac9e9530f3	fix: prompt for retry_rejected in interactive mode instead of silently applying env var RETRY_REJECTED was silently read from the environment in _configure_person, bypassing user control in interactive sessions. Now prompts the user with the env var value as the default, so the setting is visible and overridable. All other env vars in the interactive path are already correct: YEARS_FILTER is a prompt default, ONLY_PEOPLE/SKIP_PEOPLE/MIN_FACE_COUNT are auto_configure-only, and TRAINING_MODE/STRATEGY/OBJECT_CLASS are always prompted. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:03:16 +00:00
flanandClaude Sonnet 4.6	015db63d19	feat: empty CRON_SCHEDULE keeps container alive for manual docker exec Three container lifetime modes via CRON_SCHEDULE: unset — run once on startup, exit empty string — sleep infinity; use docker exec -it winnow winnow cron expression — run on startup, then on schedule This replaces the need for a separate MANUAL_MODE env var. The empty string is a natural "I want the container alive but unscheduled" signal. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 15:01:00 +00:00
flanandClaude Sonnet 4.6	2804b21f9e	refactor: auto-detect headless mode via TTY; AUTO_MODE becomes an override The primary use case is headless Docker, so auto mode is now the default whenever stdin has no TTY. Interactive mode activates when a terminal is present (docker run -it, local shell). AUTO_MODE=true remains as an explicit override for scripting with a pseudo-TTY. Removes AUTO_MODE=true, stdin_open, tty, and FORCE_CPU=false from compose.yml — none are needed for headless operation. Updates README and the interactive-mode hint in the CLI. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 14:59:09 +00:00
flanandClaude Sonnet 4.6	981a86f28a	fix: register all nvidia pip lib dirs with ldconfig; improve GPU warnings The static LD_LIBRARY_PATH only covered cudnn and cuda_runtime — missing cublas, cufft, curand, cusolver, cusparse, nvjitlink, etc. onnxruntime-gpu needs libcublasLt.so at minimum, so GPU mode silently fell back to CPU. Replace with a one-shot ldconfig call over every nvidia site-packages lib/ dir, which covers all packages regardless of what gets installed. Also: remove the ambiguous directory="" from preload_dlls (use auto-search default) and add a clear warning when CUDAExecutionProvider is absent so the user sees actionable guidance instead of silent CPU fallback. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 14:54:31 +00:00
flanandClaude Sonnet 4.6	ee585d4bae	feat: add :cpu tag for amd64 CPU-only image Introduces a VARIANT=gpu\|cpu build arg to the Dockerfile. The cpu variant uses ubuntu:22.04 (no CUDA base), installs torch+cpu and onnxruntime (no GPU deps) via a separate pyproject-cpu.toml / uv-cpu.lock, and is published as :cpu (dev-cpu on the dev branch) via a new build-cpu CI job. Saves ~2 GB over the default GPU image. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 14:50:12 +00:00
flanandClaude Sonnet 4.6	99185c2da9	fix: auto-detect missing TTY and fall back to auto mode When stdin has no TTY (Docker without -it), IntPrompt/Confirm raise EOFError and crash the container into a restart loop. Treat a non-TTY stdin the same as AUTO_MODE=true so headless runs work without any env var configuration. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 14:48:46 +00:00
flanandClaude Sonnet 4.6	196b0a5147	fix: INSIGHTFACE_HOME path, entrypoint, caching default, CRON_SCHEDULE opt-in - Dockerfile: ENV INSIGHTFACE_HOME=/models → /models/.insightface to match compose.yml and .env.example; the old value caused InsightFace to store models at /models/models/buffalo_l (double-appended subdir) - entrypoint.sh: use /app/.venv/bin/winnow (installed entry point) instead of python -m winnow.cli - config.py: ENABLE_CACHE default false → true; embedding cache is always beneficial in practice; users can opt out with ENABLE_CACHE=false - compose.yml: comment out CRON_SCHEDULE so scheduling is opt-in; flip ENABLE_CACHE to commented opt-out to reflect new default - README.md: update ENABLE_CACHE default documentation to true - tests/test_config.py: update default assertion to match Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 05:36:47 +00:00
flanandClaude Sonnet 4.6	125ce54c7f	fix: stale __version__, scheduler import order, quality test coverage - __init__.py: derive __version__ from importlib.metadata instead of a hardcoded "0.1.0" that was six releases out of date - scheduler.py: move winnow.cli import to module top (no more noqa); clean up redundant bool variables in check_models - tests/test_quality.py: 17 tests covering all five quality check functions individually plus assess_quality integration cases Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 05:33:38 +00:00
flanandClaude Sonnet 4.6	f50d3fe1ab	refactor: rename logging.py, in-process scheduler, trim .gitignore - winnow/logging.py → winnow/log_config.py: avoids shadowing the stdlib logging module; log file renamed from immich_export.log to winnow.log - scheduler.py: run main() in-process instead of subprocess.run so InsightFace and SigLIP models stay resident in memory across scheduled runs (hundreds of MB load, previously reloaded every run) - .gitignore: replaced 200-line boilerplate with ~30 project-relevant patterns; removed Django/Flask/Redis/RabbitMQ/Scrapy/etc. noise - .python-version: untracked (redundant with requires-python in pyproject.toml; kept in .gitignore for local pyenv users) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 05:30:37 +00:00
flanandClaude Sonnet 4.6	af92fe5fcc	fix: bugs and code quality in embeddings, jobs, diversity, config - embeddings: initialize ctx_id=-1 before try block so the except handler cannot NameError; move insightface_home out of try for the same reason - embeddings: replace contextlib.redirect_stdout/stderr (Python-level only) with fd-level dup2 suppression — actually silences C extension noise from InsightFace during model loading - jobs: fix frigate_count==0 falling through `or` chain; use explicit `is not None` check so a real zero is not treated as missing data - diversity: thread person_id through select_diverse_assets → _select_by_embedding → _get_face_bbox / _get_face_confidence / _crop_face_from_thumbnail so group-photo assets embed the target person's face rather than whichever person is listed first - config: replace type() hack for ConfigManager with a proper class Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 05:24:35 +00:00
flanandClaude Sonnet 4.6	4cac3d28d2	fix: onnxruntime-gpu is x86_64-only; use onnxruntime on arm64 onnxruntime-gpu has no arm64 wheels (manylinux_2_27_x86_64 / manylinux_2_28_x86_64 only). uv sync --frozen failed on the arm64 image with exit code 2. Gated onnxruntime-gpu behind the x86_64 marker; arm64 and non-Linux use the CPU onnxruntime package. Added required-environments so the lockfile is solved for both platforms. Removed onnxruntime-gpu from override-dependencies (it had no marker support and blocked arm64 resolution). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 05:05:25 +00:00
flanandClaude Sonnet 4.6	21208f8331	docs: update Python version to 3.13 in README Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-06-12 04:51:08 +00:00