Repository navigation
feat(vlm): optional detector hint for VLM labeling; remove the registry prior (closes #61) - #202
Merged
Merged
Conversation
…ctor hint The prior never lists denylisted names, ranks only classes with validated items (skipped when none), and feeds back a pending proposal only with at least 3 crops behind it. The shipped denylist gains abstract*, *_background, *_object and related words. New pack setting detector_hint_min_confidence_pct (0 = off) names the item's stored detector class in the prompt. Refs #193
…run noise words Refs #193
… item 2 The prior measured no gain on the live COCO oracle (a no-op unseeded, about -1 point and more pending proposals when seeded). Removes registry_prior_top_k, the ranking module, the source module, the 503 path, the schema row and the docs. The detector hint raised COCO-oracle overall accuracy 0.775 -> 0.853. Refs #61, #193
davidamacey
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #61. Refs #193.
The registry prior was measured on the live COCO oracle and removed; the detector hint is #61 item 2 and meets its acceptance (overall accuracy improves with it on versus off).
Numbers (gemma-4-e4b local VLM, 200 COCO images, oracle = COCO GT at IoU>=0.5, packs cloned from generic_item_v1, labels reset between runs)
Hint: overall 0.775 -> 0.853 (0.7846 -> 0.8478 in the seeded arm). Run-to-run noise is about 0.5 point.
Caveat: the detector is COCO-trained and the oracle is COCO ground truth, so the gain may be optimistic for other domains.
Registry prior (removed): unseeded it is a no-op (0.7755 off vs 0.7737 on); with ten validated classes seeded it was 0.7846 off vs 0.7747 on and pending proposals rose 85 -> 120. Pre-fix it regressed 0.7904 -> 0.7607.
Changes
detector_hint_min_confidence_pctpack setting (int 0..100, 0 = off);label_batchand the auto-label pipeline attach the storeddetector_class_nameand confidence; the prompt gets one hint line per image.registry_prior_top_k,MAX_REGISTRY_PRIOR_TOP_K,registry_prior.py,registry_prior_source.py, the 503 / stage-error prior path, the schema row, contracts, frontend fixture/pins and the related docs.pack_field_out_of_rangestays (the hint uses it).DEFAULT_PROPOSAL_DENYLIST(abstract*, _background, _object, scene, empty, blank_*, _image, shadow, ...) shared by both built-in packs and the example pack, tested against the COCO class names.scripts/curation/bakeoff/vlm_hint_oracle.py(dump/compare, with the async-bind fix).