Hello,
I could use the models you reported in the examples, like Unet + swin_tiny_patch4_window7_224 (or similar). But when I try to change the decoder (like DeepLabV3Plus) it is really tough to understand how to set the parameters.
The same foe encoders. If I try to use deit3_large_patch16_224, it is really difficult to understand how to make it works.
@bach05 I will add an example to the readme. It's a bit tricky with deeplab because it requires the encoder features to be a specific size with output_stride=8. This requires setting the scale_factors properly.
Hello, I could use the models you reported in the examples, like Unet + swin_tiny_patch4_window7_224 (or similar). But when I try to change the decoder (like DeepLabV3Plus) it is really tough to understand how to set the parameters.
The same foe encoders. If I try to use deit3_large_patch16_224, it is really difficult to understand how to make it works.
Could you please provide more example or a guide?