loss about transenet HOT 6 CLOSED

The loss.pt file is in the save folder of the experiment dir.

from transenet.

experimen

But there is no loss.pt in my experiment dir. How is loss.pt generated and saved in the experiment directory?

from transenet.

When you use this code to train on a scale factor of x4, the save folder 'TRANSENETx4_UCMerced' will be generated in the experiment directory:

python demo_train.py --model=TRANSENET --dataset=UCMerced --scale=4 --patch_size=192 --ext=img --save=TRANSENETx4_UCMerced

And the loss.pt will be saved in the 'TRANSENETx4_UCMerced' folder during training. The save code is at trainer.loss.save(self.dir)

This is the content in my generated folder here:

from transenet.

But when I run it as you did, it shows ImportError: DLL load failed while importing ft2font: 找不到指定的模块。Are there any other directories in the code that need to be modified?

from transenet.

Please tell me how to solve the error einops.EinopsError: Error while processing rearrange-reduction pattern "b (h w) (p1 p2 c) -> b c (h p1) (w p2)".
Input tensor shape: torch.Size([16, 1024, 1024]). Additional info: {'h': 24, 'p1': 8, 'p2': 8}.
Shape mismatch, can't divide axis of length 1024 in chunks of 24 after running halfway

from transenet.

Please tell me how to solve the error einops.EinopsError: Error while processing rearrange-reduction pattern "b (h w) (p1 p2 c) -> b c (h p1) (w p2)". Input tensor shape: torch.Size([16, 1024, 1024]). Additional info: {'h': 24, 'p1': 8, 'p2': 8}. Shape mismatch, can't divide axis of length 1024 in chunks of 24 after running halfway

It is suggested to check the shape of input when using rearrange function.

from transenet.