Updating onnx example to resize image to model input shape.

[kylekam]

Small PR. ONNX example currently doesn't work unless the image already matches the model input shape. Resizing allows any 3 channel image to be used.

[edgarriba]

Have you actually tried ? you can use one of these models: https://huggingface.co/kornia/ONNX_models/tree/main/models

can you also update the readme in the example exaplaining how and where to download the model ot use

[kylekam]

Have you actually tried ? you can use one of these models: https://huggingface.co/kornia/ONNX_models/tree/main/models

can you also update the readme in the example exaplaining how and where to download the model ot use

Ah okay, I didn't realize there was a huggingface location for the models. I confirmed this issue using RT-DETR from their official repo, but if the intended use case is to only use kornia models, then I can just update the readme and remove the other changes.

[edgarriba]

did it work at all with theirs ? our version provides some pre/post processings

[kylekam]

did it work at all with theirs ? our version provides some pre/post processings

Initially there was a shape mismatch, but adding the resizing seemed to fix the input issue. As for post processing, I think stuff like NMS needs to be implemented as well, so maybe it's best to recommend that people use the kornia models.

[kylekam]

Reverted the changes, linked kornia ONNX, and updated the README for clarification.

[edgarriba]

LGTM. Will merge as soon as we fix the ci