Closed zkytony closed 3 years ago
Hi,
Do you succeed to train their models?
I tried to train their SAVN model, but I might have been stuck in the environments settings (e.g. python, torch, torchvision versions). I always met some errors when I ran their codes for training.
Yes, I was able to run training and evaluation scripts for savn. But I didn't end up using this method as a baseline. I found it difficult to get this to work on the newer version of Ai2thor (3.3.4), because this codebase requires precomputed resnet features. The existing trained model (trained on ai2thor 1.0.0) doesn't work well with images from the new version.
Would you mind giving me the package versions you use to train their models? (maybe you can run conda list or pip list to obtain the information)
It is impossible to install torch==0.4.1 now, and I install PyTorch using: conda install pytorch==1.8.1 torchvision==0.9.1 torchaudio==0.8.1 cudatoolkit=10.2 -c pytorch. However I met some errors.
This is I believe the state of pytorch packages when I was working on this:
$ pip list | grep torch
efficientnet-pytorch 0.6.3
pytorch-lightning 1.3.8
segmentation-models-pytorch 0.2.0
torch 1.9.0+cu111
torchaudio 0.9.0
torchmetrics 0.4.1
torchvision 0.10.0+cu111
Thanks for your information! I have run their codes successfully now.
@YunlianMoon Glad it helped!
I just set up this repository following the README docs. The evaluation scripts for Scene Priors and Non-Adaptive A3C worked, but not for SAVN. Because pytorch 0.4.1 is out of date, I tried with 1.8.1. I needed to change
nn._VF.lstm_cell
totorch._VF.lstm_cell
. Then it worked. Just posting this here in case anyone runs into this issue.The following code also works (replace the
torch._VF
call: