fix(image): respect do_resize=False in ConvNeXT preprocessing

[RAMZI0TO99]

The ConvNeXT branch adds resize and its coupled crop despite do_resize=False. For a patterned 96x160 input with target 224, the output becomes (3, 224, 224) rather than (3, 96, 160). Already-sized inputs can also have their pixels changed.

This skips ConvNeXT geometry when resizing is explicitly disabled. An omitted flag still defaults to enabled, and subsequent rescaling/normalization continue. Tests check shape and pixel preservation for portrait/landscape/already-sized inputs, targets 224/384/512, enabled and omitted flags, subsequent processing, and unchanged CLIP/SigLIP behavior.

Validation on Windows 11, Python 3.13.5, NumPy 2.3.5, FastEmbed 0.8.1:

• 36 selected tests passed: python -m pytest -q tests/test_convnext_do_resize.py tests/test_image_transform.py.
• Local diff-scoped AST docstring coverage: 8/8 functions (100%, above the reported 80% threshold). CodeRabbit's own result is separate.
• Ruff 0.3.4 lint/format and staged whitespace checks passed.
• Repository mypy and pyright checks passed on Python 3.13.5.

The full model/OS/Python CI matrix was not run locally.

Split from #766 following the maintainer's request for one PR per problem. This branch is based directly on current main and contains only this problem's production change and regression module.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

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: 499a9dcd-1aa5-4634-9726-58a1cbfd4d3c

📥 Commits

Reviewing files that changed from the base of the PR and between 02a1aad and 97e6734.

📒 Files selected for processing (1)

• fastembed/image/transform/operators.py

🚧 Files skipped from review as they are similar to previous changes (1)

• fastembed/image/transform/operators.py

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

---

📝 Walkthrough

Walkthrough

ConvNextFeatureExtractor now defaults do_resize to true. When do_resize is false, it returns without adding resize or center-crop transforms.

Priority: ➖ Normal

Estimated code review effort: 1 (Trivial) | ~5 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 97e67

The resize flag now has the intended behavior. No actionable issue was established that would prevent merging after normal checks.

Security Architecture Review

Security architecture risk: 🔵 Low · up to d3c40

The fix preserves default behavior and makes explicitly disabled resizing work as requested. Its security significance is limited to configurations that disable resizing, where caller-supplied image dimensions can now reach inference. No security finding was established, but downstream size limits and resource exposure remain unverified.

Retained concerns
No architecture-level concerns identified.

Security review details

Security Blast Radius

• inferred — Changed dimension reachability is conditional on a ConvNeXT configuration explicitly disabling resizing. Such configurations can now deliver caller-supplied dimensions to array construction and inference unless other transforms constrain them. Whether that increases service-level resource exposure depends on unavailable model and deployment controls.

Trust Boundaries and Controls

• observed — Resize policy is read from the model directory's configuration file, whereas images enter through the embedding call. The inspected change honors that existing policy; it does not add a per-image resize-policy override. The omitted-flag path remains resize-enabled.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the change: ConvNeXT preprocessing now respects do_resize=False.
Description check	✅ Passed	The description explains the ConvNeXT resize behavior, the intended fix, and the reported tests and validation. It is directly related to the changeset.
Docstring Coverage	✅ Passed	Docstring coverage is 87.50% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 8 functions across 2 files.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ 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

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

[joein]

Thanks @RAMZI0TO99! This is a good change for future models, since it makes ConvNeXT preprocessing respect do_resize the way HF does. It doesn't affect any of our current models.
I dropped the test module since the fix is a small guard.