Problem in openslide.read_region about wsi-finetuning

[invoker-LL]

The coords should be (x,y) indeed, can you debug the changes of the variable 'self.coords_new'?

[bryanwong17]

Here are the outputs of 'self.coords_new':

[bryanwong17]

After checking again, I found out that:

topk_indexes = torch.topk(A.squeeze(), min(args.topk_num, A.shape[0]), sorted=True)[1].cpu()

The dimension of A is always [1, 1024], which makes A.shape[0] always 1. When the code executes min(args.topk_num, A.shape[0]), it proceeds with min(512, 1), resulting in the topk_indices only getting 1 index (e.g., tensor([787])).

So, from my understanding, when getting the top indices, the condition should be min(args.topk_num, A.shape[1]). Please let me know if it is correct. Thank you!

[invoker-LL]

You can try A.shape[1].
Maybe there is some error in the code, and I will check it and inform you later.