Repository navigation
feat(tracking): preserve source-reviewed depictions outside person tracks - #134
Draft
austinjiann wants to merge 2 commits into
Draft
austinjiann wants to merge 2 commits into
austinjiann wants to merge 2 commits into
Conversation
This was referenced Nov 21, 2026
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Person detectors also recognize people printed on signs. Crop recovery would spend inference on those detections and could turn them into avatars. This adds explicit, source-bound reviewed exclusions for depictions and statues, stacked on #133.
Reviews bind the original video hash/dimensions, exact sampled source frame, box, semantic reason, evidence reference and reviewer kind. Unreviewed instances remain eligible. Unique detector correspondence at IoU >0.95 is mandatory; stale or ambiguous reviews fail. Excluded masks still compete in pose association, preventing reassignment to neighbors. The exact review and counts are retained, and its content affects the paid-attempt fingerprint. Video/sampling validation runs before paid admission and submission.
Local city replay: visually inspected 61 candidate observations, excluded only 16 clear advertisement depictions, preserved all94 main-walker matches and186 unresolved observations. The elevated dark figure remains ambiguous after independent review and is not excluded. These are explicit agent annotations, not human approval or all-person acceptance.
Validation:159 focused tests passed after review fixes; Ruff and format checks pass. Full Python suite passed:2,710 tests, collected before the final schedule-validation additions; the159 focused checks include those additions. Actual model trial completed: all16reviewed advertisement observations were excluded as intended;94main-walker samples and unique assignments preserved. Crop recovery from #133 did not create an additional person track: five supported poses remained below the unchanged track-start score.181unmatched observations remain, and source overlays still miss real pedestrians. All-person acceptance failed; no avatar/clean/world stages ran. Keep the stack draft pending a viable recovery approach.
Worker-reported compute estimate:$0.04697, not an invoice; final ordinary tracking attempt3of3 consumed, with ledger history preserved. Runtime commit9eae5a3; documentation followup records the measured dead end. No automatic semantic exclusions, retry-limit changes or visual-quality promotion.