PWC-Net
PWC-Net copied to clipboard
NVlabs
forward pass
Hi, Forward pass (test) of one pair image on GPU (Titan V) takes too much time (5 min). Is it related to code or Cuda does not work well. Another point is that can I re train this model using my own data?
We haven't tested on TitanV or higher version of CUDA. But the inference time takes less than a second, excluding the model loading time. Can you check the speed for file I/O?
We didn't provide the training code in pytorch, but you can always try using your own data
Ok thanks it solved. Can I re train the model using my own data?
I have the same problem, could you tell me how to solve it? Thanks.
yes I retrained the model by my own data and I freeze some layers. tell me the problem I will help you.
On Sat, Sep 21, 2019 at 2:29 AM Zikun [email protected] wrote:
Ok thanks it solved. Can I re train the model using my own data?
I have the same problem, could you tell me how to solve it? Thanks.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/NVlabs/PWC-Net/issues/73?email_source=notifications&email_token=AIUUQRZ3UDZQLIXLGZBP6ZLQKW5LJA5CNFSM4HCCBKGKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7ILTDI#issuecomment-533772685, or mute the thread https://github.com/notifications/unsubscribe-auth/AIUUQR7KJLGZBX64AH3Z2X3QKW5LJANCNFSM4HCCBKGA .
Hi, Forward pass (test) of one pair image on GPU (Titan V) takes too much time (5 min). Is it related to code or Cuda does not work well. Another point is that can I re train this model using my own data?
I also find that the forward pass of one pair image on GPU (RTX2080) takes a few minutes （Pytorch version）. How did you solve this problem?
yes you can re train the model with your own data. The forward pass does not take much time. for me (Titan V) it took 30 ms. your code probably is working on CPU.
On Sun, Sep 22, 2019 at 12:16 AM Zikun [email protected] wrote:
Hi, Forward pass (test) of one pair image on GPU (Titan V) takes too much time (5 min). Is it related to code or Cuda does not work well. Another point is that can I re train this model using my own data?
I also find that the forward pass of one pair image on GPU (RTX2080) takes a few minutes （Pytorch version）. How did you solve this problem?
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/NVlabs/PWC-Net/issues/73?email_source=notifications&email_token=AIUUQR4PQQMWM32ISJ5PUMDQK3WRDA5CNFSM4HCCBKGKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD7I6EJY#issuecomment-533848615, or mute the thread https://github.com/notifications/unsubscribe-auth/AIUUQR54SUMFQKAY5VVO3Q3QK3WRDANCNFSM4HCCBKGA .