Some questions about train new dataset

[BloodLemonS]

Hi, I would like to inquire about the rgb images I read while training my dataset. The rgb images saved during the training process all have a gray layer, but the output was normal during testing. Also, I would like to ask about the crop_size in the code. What is the specific function of size? Can it be removed. In the process of data processing, after adjusting the image size, the problem of the tensor size will occur during training.

[youmi-zym]

Hi,

Sorry, I didn't meet this issue before, I suggest you print the shape of all tensors before saving them to the folder.

I would like to inquire about the rgb images I read while training my dataset. The rgb images saved during the training process all have a gray layer, but the output was normal during testing.

For the crop_size, you can modify it by passing a specific number to --crop_size. For more details, let's take the nyu dataset as an example, you can check it in this file.