[dajiaohuang]

Impact: Bug fix. ### Description: Fixes #3265. Persist metrics_format and restore evaluator arguments from object state, removing bare local references that fail during validation or test dataset unpickling. Validation: git diff --check passed. No tests or builds were run per assignment. ### Things to be aware of: The fix is scoped to the reported call path and preserves the existing API. ### Things to worry about: Tests and runtime validation were not run. ### Additional Context: No additional dependency or migration changes.

Summary by CodeRabbit

• Bug Fixes
  • Validation and test evaluators now retain the dataset’s restored configuration when recreated after loading a saved dataset.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 9b3ccfaf-59be-4208-bcbb-15e1d8aa2021

📥 Commits

Reviewing files that changed from the base of the PR and between f432041 and 347ac5a.

📒 Files selected for processing (1)

• python/deeplake/integrations/mmdet/mmdet_dataset_.py

Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 5 remain after this review.

---

📝 Walkthrough

Walkthrough

MMDetDataset now includes metrics_format in its pickled state. When restoring validation or test mode, it uses restored instance attributes to reconstruct the evaluator.

Changes

MMDet dataset unpickling

Layer / File(s)	Summary
Serialize state and rebuild evaluator python/deeplake/integrations/mmdet/mmdet_dataset_.py	metrics_format is saved and restored. Validation/test evaluator reconstruction uses self.bbox_info, self.pipeline, self.metrics_format, and self.num_gpus.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~8 minutes

Change: Bug fix · Severity of issue fixed: Medium

Merge Risk: ⚪ Minimal · up to 347ac

Validation and test dataset reconstruction now uses restored state, with no supported workflow failure established. The change appears mergeable subject to normal checks.

🚥 Pre-merge checks | ✅ 6 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 66.67% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 3 functions across 1 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (6 passed)

Check name	Status	Explanation
Description check	✅ Passed	The description includes all required template sections, identifies the bug fix and issue, explains the technical change, and records validation limitations.
Linked Issues check	✅ Passed	The description references and closes issue #3265, which matches the reported unpickling failure and the implemented fix.
Out of Scope Changes check	✅ Passed	The changes are limited to the affected MMDetection dataset restoration path and preserve the existing API, with no unrelated dependency or migration changes reported.
Title check	✅ Passed	The title clearly identifies the main change: restoring MMDetection evaluation datasets from pickled state.
Linked Issues check	✅ Passed	The change addresses issue #3265. It persists metrics_format in object state. It restores self.bbox_info, self.pipeline, and self.num_gpus during validation and test dataset unpickling. It rem…
Out of Scope Changes check	✅ Passed	The available whole-PR summary shows changes only in python/deeplake/integrations/mmdet/mmdet_dataset_.py. The changes support issue #3265 and preserve the public API. No unrelated dependency, migra…

• Fix all pre-merge checks with AI

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{I’m a rabbit with a dataset to pack,
I tuck metrics_format in the stack.
Restored fields guide the evaluator’s start,
With bbox and pipeline each playing their part.
I hop off with state all intact.}

_{Comment @coderabbitai help to get the list of available commands.}