[WIP] Experiment with different input image resizing sizes

[Copilot]

Plan to replace hardcoded resize strategy with model configuration value:

• Understand the current codebase and identify where resize strategy is hardcoded
• Modify Classifier class to expose the resize_strategy from model config
• Update app.py to use the resize strategy from the classifier instead of hardcoded value
• Test the changes to ensure they work correctly
• Run code review and security checks

Original prompt

---

This section details on the original issue you should resolve

<issue_title>experiment with different resizing size</issue_title>
<issue_description>Currently we're using the default values for input image resize, ranging around 230 (see resize_size values in https://pytorch.org/vision/main/_modules/torchvision/models/convnext.html). However the paper (https://arxiv.org/pdf/2201.03545) reports in table 1 that a higher resolution generally leads to better performance. We'd like to see if the case hold for our data as well.
</issue_description>

<agent_instructions>this is simply replacing redundant variable definition with a single source of truth. </agent_instructions>

Comments on the Issue (you are @copilot in this section)

@keighrim v8.0 development included experiment with three different preprocessors https://github.com/clamsproject/app-swt-detection/blob/c4f516482aac7b02394e6af673bb350b7b7c3eee/modeling/data_loader.py#L50-L66

and multiple rounds of grid searches reported distorted is empirically the best choice, hence the v8.0 model was released after trained with distorted training document, and the inference also use the same resizing strategy

app-swt-detection/app.py

Lines 122 to 123 in c4f5164

	resizer = ImageResizeStrategy.get_transform_function('distorted')
	resized = torch.stack(tuple(map(resizer, extracted)))

Looking at the code now, I think we can do better by not hardcoding the resizer name in the app.py . This issue will be closed when app.py is updated and automatically decide the resizer name based on model configuration. </comment_new>

• Fixes experiment with different resizing size #130

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.