Input dimension limitation

[tawsinDOTuddin]

The code seems to be only work on a specific input resolution which the model trained on. If I want to test it with dimension other than 432x240, I can but I have to train the model with that specific dimension. right?

[ruiliu-ai]

Correct.
If you want to test your model with arbitrary-size input, you can remove the positional embedding layer. That's the key reason why the input dimension must be fixed.

[tawsinDOTuddin]

I think it is same issue with your another work "Decoupled Spatial-Temporal Transformer for Video Inpainting"....what i should do if i test that model with arbitrary-size input?

[ruiliu-ai]

Remove positional embedding, and edit this line to be 1/4 * (h, w) where (h, w) can be arbitrary input/output image size
https://github.com/ruiliu-ai/FuseFormer/blob/master/model/fuseformer.py#L119

[ruiliu-ai]

Please refer to ruiliu-ai/DSTT#4

[tawsinDOTuddin]

Thank you for your response :)