Closed darthandvader closed 4 months ago
This looks like the features stored in the npy files were not formatted properly. The THUMOS dataloader expects video features of size T x C in each npy file, with T as the number of time steps and C as the number of feature channels. You might want to double check this part.
Closed due to inactivity.
File "/actionformer_release/libs/modeling/meta_archs.py", line 409, in preprocessing padfeat[..., :feat.shape[-1]].copy(feat) RuntimeError: The size of tensor a (3) must match the size of tensor b (360) at non-singleton dimension 2
Hi, I formulate my dataset in the format of thumos and used thumos config to train, but encountered this error, wondering how to address that.