Feat/luxonis eval object

[klemen1999]

Purpose

Summary

• Introduces a reusable LuxonisEval evaluation object so setup, evaluation, and cleanup are owned by a Python API instead of living entirely inside the CLI flow.
• Keeps the CLI path intact by turning eval_run into a thin wrapper around the new object-based lifecycle.
• Hardens engine/resource lifecycle handling and fixes class-index mapping/report rendering issues that could affect metric correctness.

What Changed

• Adds luxonis_eval/core.py with setup(), evaluate(), and close() methods, runtime component creation, compatibility checks, and structured result output.
• Refactors luxonis_eval/__main__.py so eval_run(...) can accept config inputs directly, returns results, and always closes resources in a finally block.
• Exports LuxonisEval from luxonis_eval/__init__.py and updates the README Python API/docs to cover both one-shot usage and explicit lifecycle control.
• Refactors BaseEngine to use framework-managed setup() plus subclass _setup_impl(), and replaces teardown() with close().
• Updates the DepthAI engine and ONNX engine to support explicit setup/close, guard inference before setup, and clear runtime metadata on close.
• Replaces the old LuxonisLoader monkey-patching path with resolve_luxonis_loader_class_mapping() and builds metric contexts once during evaluator setup.
• Fixes segmentation/classification metrics so class_index_map is only applied when it actually exists, avoiding incorrect remapping in DiceCoefficient, MIoU, and TopKAccuracy.
• Fixes report generation in make_report_table() so rendering no longer mutates metric result dictionaries by removing the "metric" key.

Migration Notes

• Custom engine implementations now need to implement _setup_impl() and close() instead of the previous setup() / teardown() contract.
• Programmatic callers can continue using eval_run(...), but it now returns a structured results dictionary instead of being CLI-only.

Specification

None / not applicable

Dependencies & Potential Impact

None / not applicable

Deployment Plan

None / not applicable

Testing & Validation

None / not applicable

AI Usage

Assisted-by: AGENT_NAME: Codex

Submitted code was reviewed by a human: YES

The author is taking the responsibility for the contribution: YES

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: a08df7f1-f1c0-443b-a79f-0c398006c4dc

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/luxonis_eval_object

---

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

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