lama
lama copied to clipboard
Dataset Dataloader to load the label image at the same time
Hi, i would like to use the network as backbone for segmentation. And i want to load the label image as well. So i wrote some lines as the image but i caught problem. The code can run for just 1 epoch, by the 2. epoch it shows error in image.
Could you please give me some advice how i can solve it? Thank you so much.
The fact that the error occurs after a full epoch of training suggests me that the problem might be with the validation dataset. Does validation go well?
Hello, i think the problem occurs at train dataset, i have changed the train dataset loading according to /evaluation/data.py InpaintingDataset loadfile. The problem doesn't exit.
I still got another question. In /saicipainting/evaluation/data.py, when we load image we have divide 255 for each data image and mask image. But for training, in /saicipainting/trainning/data/dataset.py, the images are not divided by 255. Why is it?
When i load the label image by evaluating, do i also have to divide it by 255 ?