cw

Some questions about decoder position embedding for masked tokens

1

In the decoder position embedding matrix, the size of first dim is the number of patches + 1, as the 1 for ViT's cls_token. But when embedding the position for...

CUDA_ERROR_INVALID_HANDLE: invalid resource handle occured when using another gpu instead of the default card(0)

1

My server has multi gpus, when I change '.cuda()' to '.cuda(1)', exception occured: cupy.cuda.driver.CUDADriverError: CUDA_ERROR_INVALID_HANDLE: invalid resource handle I also add 'os.environ["CUDA_VISIBLE_DEVICES"] = 1', but it doesn't work. Can anyone...

About the loss computation

1

I was confused about why use BCE to compute the loss of 'x' & 'y', could it be MSE? https://github.com/BobLiu20/YOLOv3_PyTorch/blob/c6b483743598b5f64d520d81e7e5f47ba936d4c9/nets/yolo_loss.py#L55

Details about training target and loss function

1

Appreciate this excellent work! But I'm confused about the training target and loss function. According to the paper, training target is to recover the masked area of source image with...