search

SpadeLiu / Lac-GwcNet

Local Similarity Pattern and Cost Self-Reassembling for Deep Stereo Matching Networks
MIT License
44 stars 9 forks source link

Tensor pushing from GPU to CPU #8

Closed Salvatore-tech closed 2 years ago

Salvatore-tech commented 2 years ago

Good evening, I'm bit struggling with the storing place of tensors, in fact when running python KITTI_ft.py --data_path /home/s.starace/Dataset/KITTI_2015/training \ --load_path ./preTrained/SceneFlow.pth
It suppose that tensors should be pushed first on the CPU before converting them through numpy... could you confirm that? Thanks in advance tensorErrorLog.txt

SpadeLiu commented 2 years ago

Hi,

Yes, I think so. You can either move "disp_true" onto CPU after model forward process or don't push it on GPU at first and create a zero matrix which has the same shape as "disp_true" and replace "disp_true" with it as the model input. Now I am not convenient to change the code. After 11 hours, I will upload a new version.

SpadeLiu commented 2 years ago

Good evening, I'm bit struggling with the storing place of tensors, in fact when running python KITTI_ft.py --data_path /home/s.starace/Dataset/KITTI_2015/training \ --load_path ./preTrained/SceneFlow.pth It suppose that tensors should be pushed first on the CPU before converting them through numpy... could you confirm that? Thanks in advance tensorErrorLog.txt

Hi，

Could you please try the lateset code?

Salvatore-tech commented 2 years ago

Good evening, I'm bit struggling with the storing place of tensors, in fact when running python KITTI_ft.py --data_path /home/s.starace/Dataset/KITTI_2015/training \ --load_path ./preTrained/SceneFlow.pth It suppose that tensors should be pushed first on the CPU before converting them through numpy... could you confirm that? Thanks in advance tensorErrorLog.txt

Hi，

Could you please try the lateset code?

Everything is alright after db029f8 I'll close the issue, thanks again!