Closed dreamer-1996 closed 3 years ago
Hi, @dreamer-1996
They are not the same type of files - I guess.
As for the file type, they are the same. I'm not sure why you thought so, but maybe is it because you got the following error when you tried to load the model in pytorch_mpiigaze_demo
program?
ModuleNotFoundError: No module named 'gaze_estimation'
If so, sorry for the inconvenience. That error occurs when you try to open the .pth
file saved by this repo without including this repo in your PYTHONPATH
. This is because the checkpoint file contains the config
which is an instance of a class defined in this repo.
I'll fix this, but for the time being, you can work around this error by adding this repo to your PYTHONPATH
or by removing the config
key from your checkpoint file.
Thank you for the reply.
I am trying to use this lenet and resnet architecture to convert 3d angles to 2d gaze points on the screen.I have trained a model using the architecture provided by you.Now I want to use that trained model to infer the 2d gaze points in real time.Do I have to save my trained model using torch.save()
or can I use the .pth
files that are getting generated inside the experiments folder?
@dreamer-1996
As I said, if you add this repository to your PYTHONPATH
, you should be able to use checkpoint files in your experiments folder. Doesn't it work? If so, what error are you getting?
@hysts oops I forgot to close this issue. Yes it works. Thank you :)
Hello :) It's me again. I can see that there are
.pth
files getting generated as checkpoints in the experiments folder while I am training the model.I can also see
.pth
files being released by you as Version 1.0. They are not the same type of files - I guess.Also anywhere in the code provided in the repository, is there any provision to save the trained model? Or do I have to do that by myself to save a trained model so that I can use the model to infer later.