jacobian_regularization branch: Demo fails, many bugs. about atlasnet

[ThibaultGROUEIX]

Interesting let me test it.

[ThibaultGROUEIX]

You can git pull.

I've tested training both autoencoder and SVR, and fixed path and reload issue in the demo.
You can now try : python train.py --demo --demo_input YOUR_PATH.obj --reload_model_path YOUR_PATH/network.pth --template_type SQUARE

Note that the pretrained models are only autoencoders. If you want to do single-view, you have to train new models.

Best regards,
Thibault

[Howardyangyixuan]

You can git pull.

I've tested training both autoencoder and SVR, and fixed path and reload issue in the demo.
You can now try : python train.py --demo --demo_input YOUR_PATH.obj --reload_model_path YOUR_PATH/network.pth --template_type SQUARE

Note that the pretrained models are only autoencoders. If you want to do single-view, you have to train new models.

Best regards,
Thibault

Great thanks for your fast response and swift change. It seems to work well. However, it requires torchvision version 0.3.0 or later, which means you need pytorch 1.2 or later, and it recursively needs cuda 9.2 or later. Otherwise, you will meet some problem due to an old version of torchvision and have to change some code. I will try it later. But again great thanks!

[Leerw]

Are there available for more nb_primitive pretrained models other than 1 patch?

[ThibaultGROUEIX]

No, all trained models for this branch are here.
Best regards

[Leerw]

No, all trained models for this branch are here.
Best regards

Thanks.