Run pretrained models on testing data

[chiffonng]

Hi authors, thanks for your wonderful work!

I wish to run just the inference of pretrained models on testing data before running it on my own data, but I'm not sure how to specify checkpoint path for the argument --ckpt.

The models downloaded and unzipped from Google Drive do not have .pt or .pth files for saved Pytorch checkpoints.

The unzipped directory for one model has the following format:

archive
|------ data/
|---------------- 0
|---------------- 1
|---------------- ... 2k more files 
|------ data.pkl
|------ version

Given this, what path/file should I assign to --ckpt or the argument in configs.py?

Sorry if this is a beginner question! I'm new to robotics research and Pytorch in general.

[liujiuming123]

Hi, pretrained model is speficied by the parameter 'ckpt' in configs.py file. And you can replace the parameter name with the pretrained checkpoint's name we provided.

[chiffonng]

Sorry if the question isn't clear! I did change configs.py to specify ckpt. I edited my original question. Basically, I'm not sure which of these are pretrained weights

archive
|------ data/
|---------------- 0
|---------------- 1
|---------------- ... 2k more files 
|------ data.pkl
|------ version

Is it something like archive/data/0? I tried several paths already.

[chiffonng]

@liujiuming123 Is there perhaps any update on this?

[liujiuming123]

Hi! There is no need to unzip it. Just download the file with the ending of .pth.tar and copy this on --ckpt.

[chiffonng]

Thank you, the model checkpoints can be loaded now. I run into other issues but I will look into them first.