From fe4cfac7b52058c0ff3d925bdd3d2be4c800ac71 Mon Sep 17 00:00:00 2001 From: Holden Date: Sun, 14 Jun 2026 16:06:45 +0000 Subject: [PATCH] docs: add missing dedup step, fix auto-stop description in How It Works - Step 5 (near-duplicate removal) was added in v0.4.4 but never documented; added between embedding and diversity selection - Auto-stop was described as 'stops when similarity exceeds threshold' which is backwards; it stops when the next candidate's distance to already-selected images falls below the threshold (too similar, not enough new information) - Renumbered steps 5-8 to 6-9 - Tightened hard-example weighting description to match the code --- README.md | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-) diff --git a/README.md b/README.md index c9cfb26..a966efd 100644 --- a/README.md +++ b/README.md @@ -43,24 +43,30 @@ Immich library • Objects → SigLIP (Vision Transformer) → 768-dim vector │ ▼ -5. Diversity selection +5. Near-duplicate removal — greedy cosine-distance pass drops burst shots + and near-identical photos before clustering runs; the highest-quality + image from each near-duplicate group is kept + │ + ▼ +6. Diversity selection • K-Medoids clustering → one representative per natural group • Farthest Point Sampling → fill remaining slots with maximally spread picks - • Hard example weighting — unusual angles and low-confidence detections - are biased toward selection, since those are where models tend to fail - • Auto mode: stops when similarity to the existing set exceeds a threshold - (20 % of median pairwise distance for faces, 10 % for objects) + • Hard example weighting — low-confidence detections get a distance boost + so unusual angles and harder looks are preferred over easy frontals + • Auto mode: stops when the next candidate is too similar to those already + selected (distance threshold = 20 % of median pairwise distance for + faces, 10 % for objects) │ ▼ -6. Download full-resolution originals from Immich +7. Download full-resolution originals from Immich │ ▼ -7. Crop and process +8. Crop and process • Face mode: EXIF-corrected, landmark-aligned 112×112 crop (ArcFace format) • Object mode: YOLOv9c detection → one crop per matched instance │ ▼ -8. Deliver +9. Deliver • Face mode: upload crops to Frigate's face registration API ↳ below MAX_AUTO_IMAGES — upload freely ↳ at cap + QUALITY_REPLACEMENT=true — with Frigate scoring active,