viola521

when I run the train_full.py ,there is no OOM error reported. lat_features = features.expand(N,*features.size()).permute(1,0,2,3,4).contiguous()[typs] torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 2.78 GiB (GPU 0; 7.79 GiB total capacity;...

> I found this problem too. In lav_agent_fast.py, I multiplied desired_speed by 3. Seems to work better, but not sure how will it affect the infractions. > > Also I...