search

ouenal / scribblekitti

Scribble-Supervised LiDAR Semantic Segmentation, CVPR 2022 (ORAL)
https://ouenal.github.io/scribblekitti/
142 stars 17 forks source link

Batch size bigger than 1 #3

Closed l1997i closed 2 years ago

l1997i commented 2 years ago

Thanks so much for your excellent work and code. I've a question about the dataloader code. Is it possible to set the batch size bigger than 1 (e.g., 4, 8 or 16). When I tried to set a bigger batch_size in training.yaml, the code went wrong with the error "RuntimeError: stack expects each tensor to be equal size, but got [124266, 3] at entry 0 and [112695, 3] at entry 1".

# in `training.yaml`
train_dataloader:
  batch_size: 4       # default is 1
  shuffle: True
  num_workers: 4

Thanks in advance!

ouenal commented 2 years ago

It's not possible with the current state of things. I couldn't run with a batch size higher than 1 due to the 12GB gpu memory limitation that I had, so I didn't really consider higher batch sizes. I would suggest writing a new collate function for the dataloader to pad the point clouds to a common number of points. That should allow for higher batch sizes. You can also refer to the Cylinder3D codebase to see how the collate function can be implemented without padding.