0

I am experimenting with designing a semantic segmentation network using Pytorch. It performs well on my computer. For better performance, we experimented by moving the network to a computer with a lot of GPU capacity. Basically, if you set the same environment for only the versions of Pytorch and torchvision and proceed with the experiment, performance degradation occurs even on the moved PC or Google Colab. I just copied and pasted the code and ran it as a test before doing other experiments.

The network structure is the same, but are there other external factors that will degrade performance? (ex: gpu, ram, etc...)...

SeunghwanByun
  • 21
  • 1
  • 1
    Aside from Pytorch - have you checked if the CUDA versions are the same? Maybe your training/validation/testing set have been corrupted/compressed? Have you tried setting up a python venv to copy your dependencies exactly? Have you tried saving your torch weights on you local machine and checking the performance on the remote PC? – nablag Dec 11 '20 at 07:35
  • 1
    @nablag Yes, I checked that the CUDA version is the same, and moved to the training/test set in case there was a problem. I'm not sure if it's compressed. When I saved the torch weight in the local system and checked the performance on the remote PC, there was no problem. Python venv is a method I haven't tried. I should try. – SeunghwanByun Dec 11 '20 at 10:00
  • You can check if there is image compression if the size of the images on the server are smaller than the images on your local machine. – nablag Dec 11 '20 at 15:21

0 Answers0