Pre-trained model in Pytorch

Question

I just loaded the pre-trained model 3D resnet https://github.com/kenshohara/3D-ResNets-PyTorch .

I loaded this model in this way after i download from the link above the file 'resnext-101-kinetics.pth' : way i defined the model

this model work on a big dataset of videos (unfortunately i don't have enough space on my pc and i didn't find a link that i can download part of the data set). I just want to know if i load the model correctly and to know that this net is trained. There is an option to be sure that the model loaded well? I printed the state_dict and it prints the weights.. Thanks for any help.

Answer 1

It already includes the script to load the pretrained weights to the model. Please check the code from the line 168 - line 173:

https://github.com/kenshohara/3D-ResNets-PyTorch/blob/master/model.py#L168-L173

Change the value for default in the opts.py file, to the path of the file you downloaded resnext-101-kinetics.pth.

'--pretrain_path', default='path/to/resnext-101-kinetics.pth', type=str, help='Pretrained model (.pth)')

If there is no error appears when you run your code, then the pretrained file resnext-101-kinetics.pth is loaded properly.