Skip to content

chore(autoannotate): bump model to pyronear/yolov11s v8.2.0, conf 0.05 - #388

Merged
MateoLostanlen merged 1 commit into
mainfrom
chore/autoannotate-model-v8-2-0
Sep 14, 2026
Merged

MateoLostanlen merged 1 commit into
mainfrom
chore/autoannotate-model-v8-2-0

Conversation

@MateoLostanlen

Copy link
Copy Markdown
Member
  • Point the baked auto-annotate ONNX at pyronear/yolov11s, pinned to tag v8.2.0, instead of pyronear/yolo11s_sensitive-detector @ main.
  • Raise the detector confidence floor from 0.01 to 0.05, so fewer and higher-confidence boxes reach the engine-overlap gate.
  • Record the new model in AUTOANNOTATE_MODEL_NAME / _VERSION.
  • No detector code change: the tarball still extracts flat to /app/models, the ONNX IO names are still images / output0, and the image size is still 1024.

Validation: image builds (so the pinned URL resolves and extracts); inside it /app/models holds a single best.onnx of 38532985 bytes matching the release manifest, AUTOANNOTATE_CONF reads 0.05, SmokeDetector loads the model and predict() runs. Full API suite: 816 passed.

Existing auto_predictions are untouched; re-running auto-annotate on already-processed sequences is a separate ops call.

Point the baked ONNX tarball at the yolov11s release repo, pinned to tag
v8.2.0, and raise the detector confidence floor from 0.01 to 0.05. The
tarball layout (flat best.onnx), ONNX IO names and 1024 image size are
unchanged, so the detector code stays as is.
@MateoLostanlen
MateoLostanlen merged commit 5478a6e into main Sep 14, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant