Fix deploy_to_roboflow seg model export

[Matvezy]

What does this PR do?

For seg models models when calling deploy_to_roboflow and deploy, this error occurs:

File "/app/app.py", line 371, in process_versioned_model_upload
    conversions = prepare_files_and_convert(cache_dir, model_type, f"{workspace_id}/{dataset_id}/{version_id}")
  File "/app/app_rfdetr.py", line 44, in convert
    model.export(output_dir=cache_dir)
  File "/usr/local/lib/python3.10/dist-packages/rfdetr/detr.py", line 123, in export
    self.model.export(**kwargs)
  File "/usr/local/lib/python3.10/dist-packages/rfdetr/main.py", line 581, in export
    print(f"PyTorch inference output shapes - Boxes: {dets.shape}, Labels: {labels.shape}, Masks: {masks.shape}")
AttributeError: 'dict' object has no attribute 'shape'

Type of Change

• Bug Fix

Fixing this by calling .eval() on the right model object.

Testing

• I have tested this change locally

Test details:
N/A

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code where necessary, particularly in hard-to-understand areas
• My changes generate no new warnings or errors
• I have updated the documentation accordingly (if applicable)

Additional Context

[codecov]

Codecov Report

❌ Patch coverage is 0% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 22%. Comparing base (978de68) to head (3cdf0b1).

Additional details and impacted files

@@          Coverage Diff           @@
##           develop   #578   +/-   ##
======================================
  Coverage       22%    22%           
======================================
  Files           46     46           
  Lines         6181   6181           
======================================
  Hits          1342   1342           
  Misses        4839   4839           

🚀 New features to boost your workflow:

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

[Borda]

@Matvezy could we add a test, pls

[Copilot]

Pull request overview

Fixes ONNX export for segmentation models when using deploy_to_roboflow / deploy by ensuring the correct model instance is put into evaluation mode prior to inference during export.

Changes:

• Switches the .eval() call in export() from self.model to the locally-copied model used for inference/export.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[junxnone]

👍

[jrdalenberg]

This circumvents the problem in one scenario but the same bug also appears when just exporting to onxx.

The problem occurs one line before the print statement:

masks = outputs['pred_masks']

outputs['pred_masks'] is a dict and has attributes ['query_features'] and ['spatial_features']. This dict has no shape attribute, hence the error.

Depending on which needs to be printed out, the line should be updated to:

masks = outputs['pred_masks']['query_features'] # or ['spatial_features']

Then the export works successfully.