search

microsoft / transductive-vos.pytorch

a transductive approach for video object segmentation
155 stars 32 forks source link

the problems of pretrainedmodel #4

Closed pixelsmaker closed 4 years ago

pixelsmaker commented 4 years ago

I've downloaded the davis_train.tar and youtube_train.tar, but I can't successful unzip the tar package, it seems that some data is corrupted, I don't know where the problems come from...

pixelsmaker commented 4 years ago

And I've only got 2 GPUs with 11 GB CUDA memory each, how can I change the parameters to train the model on DAVIS?

eastOffice commented 4 years ago

Hello,

  1. the .tar file is the model that can be directly loaded by torch.load(), which doesn't need to be untared.
  2. You can try a smaller batchsize first, the rest of the training parameters may not be optimal for a smaller batchsize, but shouldn't be very different from the default ones.