[HuaTNA]

Description

Fixes #724.

Jina CLIP preprocessing rejects interpolation="nearest" and silently changes interpolation=0 / Image.Resampling.NEAREST to bicubic. Pillow's nearest-neighbor enum is zero, so both truthiness checks treat a valid selection as missing.

Use explicit None checks for the string lookup and numeric fallback. Other interpolation modes, the bicubic default, and rejection of invalid names are preserved.

Add pixel-level tests through Compose.from_config, comparing a resized checkerboard with Pillow's reference result. Coverage includes string and mixed-case names, zero and enum inputs, every other supported string mode, the default, and invalid names. No model download is needed for the new tests.

Validation

• Before the fix: pytest -q tests/test_image_transform.py — 4 failed, 18 passed. String nearest modes raise ValueError; integer/enum nearest modes produce incorrect pixels.
• After the fix: pytest -q tests/test_image_transform.py tests/test_preprocessor_utils.py tests/test_postprocess.py — 54 passed.
• Ruff 0.3.4 lint and format checks, including repository pre-commit hooks, pass for both changed files.
• mypy fastembed --disallow-incomplete-defs --disallow-untyped-defs --disable-error-code=import-untyped — no issues in 64 source files.
• Python 3.11.15, Pillow 12.3.0, NumPy 2.4.6, macOS arm64. The full model/GPU test suite was not run.

All Submissions

• Followed the contribution guidelines.
• Checked existing issues and PRs for the same change.
• Added regression tests and ran the relevant tests locally.
• Installed and ran the repository pre-commit hooks.

AI assistance (Codex) was used for investigation, implementation, and local test execution.

[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: 76bb82a8-3c87-4959-ae38-42f0eb98a0ac

📥 Commits

Reviewing files that changed from the base of the PR and between 8de28b8 and e69f352.

📒 Files selected for processing (2)

• fastembed/image/transform/operators.py
• tests/test_image_transform.py

Included review availability: Your plan provides up to 8 included reviews per hour; 7 remain after this review.

---

📝 Walkthrough

Walkthrough

The interpolation resolver now distinguishes an unresolved value from a valid falsy Pillow resampling value. The Jina CLIP resize fallback selects bicubic only when interpolation is None. New tests cover supported interpolation inputs, including nearest-neighbor string, integer, and enum values, and verify that empty or unknown strings raise ValueError.

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

Merge Risk: ⚪ Minimal · up to 690aa

The change preserves nearest-neighbor resizing for the reported inputs without changing the default. No merge-blocking issue is identified.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 4 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: preserving nearest interpolation for Jina CLIP.
Description check	✅ Passed	The description directly explains the interpolation bug, the fix, the regression tests, and validation results.
Linked Issues check	✅ Passed	The implementation satisfies issue #724. Compose._get_resize now preserves 0 and Image.Resampling.NEAREST and uses bicubic only when interpolation is None. _interpolation_resolver now return…
Out of Scope Changes check	✅ Passed	The changes stay within issue #724. They modify interpolation resolution and add focused regression tests for the Jina CLIP image processor. No unrelated production behavior or unrelated files changed…

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

---

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.}

[joein]

Hey @HuaTNA

Thanks for pointing it out and fixing this!

[joein]

Though, I decided to remove the tests, since this path has actually never been hit. Jina's model uses bicubic, and it's the only model which uses this processing path.