Can we test on arbitrary size images? about deep-image-matting HOT 4 CLOSED

@oee5215, the pytorch version of deep-image-matting (https://github.com/huochaitiantang/pytorch-deep-image-matting) is preferred and yields better performance.

from deep-image-matting.

We cannot.

from deep-image-matting.

@oee5215, just modify the shape of the input tensor to be (None, None, 4) and re-train the model. In this condition, the model can handle inputs with arbitrary sizes.

Edit:
After following the above suggestions, it seems that this method does not work. I will test other options and see if it can use inputs with arbitrary sizes.

from deep-image-matting.

@cardwing Thank you! Please let me know if you find a working solution, since it would be ideal to use this model for my research project.

from deep-image-matting.