utils.MLPCenter issue

[zjz82108008]

utils.MLPCenter block seems to be something wrong.
The output of the model was not used in the test section:

                scores = torch.matmul(acts, patterns) # 512, num_blocks, vary from 0 to 1
                mask, labels = torch.topk(scores, k=int(num_blocks*0.2), dim=-1)
                mask = mask > 0
                pred = model(input)
                pred = torch.topk(pred, k=int(num_blocks*0.2), dim=-1)[1]

                for x, m, s in zip(labels, mask, scores):
                    if m.sum().item() == 0:
                        continue
                    x = sum([s[xx] for xx in x.cpu()]).item()
                    y = s.sum().item()
                    acc.append( x / y)
                cur_acc = np.mean(acc)

[zzy14]

Hi, thanks for pointing out this issue. labels, mask, scores should be pred, mask, scores instead. It has been solved.