[tongwentao]

Description

In-memory dataset exports ignore a False result from cv2.imwrite. An export can therefore return normally even though an image was never saved. Both the shared detection-image exporter and classification folder exporter now raise OSError containing the destination path, matching ImageSink.

Minimal reproduction (the destination image path already exists as a directory):

from pathlib import Path
from tempfile import TemporaryDirectory

import numpy as np
import supervision as sv

with TemporaryDirectory() as directory:
    output = Path(directory)
    (output / "image.png").mkdir()
    dataset = sv.DetectionDataset(
        classes=[],
        images={"image.png": np.zeros((4, 4, 3), dtype=np.uint8)},
        annotations={"image.png": sv.Detections.empty()},
    )
    dataset.as_yolo(images_directory_path=str(output))

Before: returns None, with no image file written. After: raises OSError: Failed to save image to path: .../image.png.

Motivation and Context

Report incomplete exports at the point of failure, so users can correct the destination instead of discovering missing training images later.

Changes Made

• Check image-write results in save_dataset_images and ClassificationDataset.as_folder_structure, and document the exception.
• Add real filesystem failure cases and successful pixel-preservation tests for both exporters; add an Unreleased changelog entry.

Testing

Revalidated after merging current upstream develop (893a9b1a) and preserving the fix under Unreleased.

• New regression cases against unmodified develop: two failed with DID NOT RAISE OSError; both successful-write controls passed.

• MPLBACKEND=Agg .venv/bin/pytest src/ tests/ --cov=supervision --cov-report=term: 6260 passed, 1 skipped, 44 warnings, 94% coverage. This includes the integration test running the ordinary suite with OpenCV imports blocked.

• .venv/bin/pre-commit run --all-files: all configured pre-commit hooks passed, including Ruff, docformatter, mypy and codespell.

• Tested on Python 3.12 with OpenCV headless 4.14.0.94.

• Added/updated tests, and the full suite passes locally

• Updated docs (docstrings / mkdocs entry) for new or changed public API

• Added a changelog entry in docs/changelog.md under Unreleased

[Copilot]

🟢 Approval recommended

Both narrowly scoped failure checks have focused regression coverage and matching documentation, with no unresolved findings.

No open issues 🎉

What changed in this PR

🟢 Approve — This change prevents silent image-write failures in Supervision’s in-memory dataset exports.

Changes:

• Raise OSError with the destination path when either exporter receives a failed write result.
• Add filesystem-failure and successful pixel-preservation tests.
• Document the exception and update the changelog.

No actionable issues identified. Behavior, error handling, tests, and documentation are covered by the supplied review evidence; tests were not rerun here.

Quality: code 5/5 (minimal checks), tests 5/5 (failure and success coverage), documentation 5/5 (matching updates). Risk: 1/5.

File	Description
tests/​dataset/​test_utils.py	Tests detection image-write failures and pixel preservation.
tests/​dataset/​test_core.py	Tests classification image-write failures and pixel preservation.
src/​supervision/​dataset/​utils.py	Checks detection image writes and documents errors.
src/​supervision/​dataset/​core.py	Checks classification image writes and documents errors.
docs/​changelog.md	Announces the export failure handling.

_{Review effort: Balanced}

[CLAassistant]

All committers have signed the CLA.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94%. Comparing base (e82da41) to head (c23d3e7).

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #2772   +/-   ##
=======================================
  Coverage       94%     94%           
=======================================
  Files           83      83           
  Lines        13659   13661    +2     
=======================================
+ Hits         12863   12867    +4     
+ Misses         796     794    -2     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.