for cuda 11 pytorch 1.8 bug fix

raphaelmemmesheimer / skeleton-dml

Skeleton-DML: Deep Metric Learning for Skeleton-Based One-Shot Action Recognition

23 stars 7 forks source link

Open shinji-ohkubo opened 3 years ago

shinji-ohkubo commented 3 years ago

becomes a bug in cuda 11 pytorch 1.8 and pytorch 1.9. It will be fixed below.

train.py line 236 before: classifier = torch.nn.DataParallel(MLP([cfg.embedder.size, cfg.embedder.class_out_size])).to(device) after: classifier = torch.nn.DataParallel(MLP([cfg.embedder.size, cfg.embedder.size])).to(device)
train.py lin99 before: accuracy = accuracy_score(knn_labels, query_labels) after: accuracy = accuracy_score(knn_labels.cpu(), query_labels.cpu())

sasakits commented 2 years ago

Hi Thank you for your information,

After modify code with your suggestion, I have met this Error

TypeError: vars() argument must have __dict__ attribute

my env: RTX 3090, pytorch 1.9, cuda 11 If you know anything, please let me know

Thanks,