Skip to content

fix(eval): resolve VitPose image processor metadata - #1310

Draft
ssss141414 wants to merge 1 commit into
microsoft:mainfrom
ssss141414:vitpose-eval-processor-fallback
Draft

fix(eval): resolve VitPose image processor metadata#1310
ssss141414 wants to merge 1 commit into
microsoft:mainfrom
ssss141414:vitpose-eval-processor-fallback

Conversation

@ssss141414

Copy link
Copy Markdown
Contributor

Summary

Fixes keypoint-detection evaluation for VitPose checkpoints whose processor metadata names ViTPoseImageProcessor while Transformers exposes VitPoseImageProcessor and cannot resolve the checkpoint through AutoImageProcessor.

This VitPose-family evaluator fix was extracted from #1299 because it is independent of model build, recipes, and Perf.

Changes

  • Keep AutoImageProcessor as the primary path.
  • Fall back to VitPoseImageProcessor only for the known Unrecognized image processor failure and metadata that identifies the VitPose family.
  • Preserve trust_remote_code forwarding.
  • Fail early with a clear error when model_id is absent.
  • Do not catch unrelated processor errors or alter successful non-VitPose resolution.

Validation

  • tests/unit/eval/test_keypoint_detection_evaluator.py: 14 passed
  • Ruff check: passed
  • Ruff format check: passed
  • Previously validated real COCO one-row functional smoke processed one sample and emitted mAP=0.8; this is operability evidence, not an accuracy claim.

Split from #1299.

@ssss141414 ssss141414 added the model-scale-by-skill Model support PR created or maintained by the adding-model-support skill label Aug 14, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

model-scale-by-skill Model support PR created or maintained by the adding-model-support skill

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant