Environment: python 3.6 PyTorch 1.0.0

The code is implemented on the base of jongchyisu's mvcnn_pytorch, and is inspired by Sean Kim's tensorflow version.

The inception v4 can be found here.

The paper can be found here.

There are two main difference from the origin paper:

1. Considering the huge amount of fc layer, we use 1*1 conv instead of fc in group schema module.
2. If all views' scores are very small, it may cause some problem in params' update. So we add a softmax to normalize the scores generate by group schema.

modify root in data_process.py to your ModelNet40 dataset.

python data_process.py

python gvcnn_train.py

• Feng Y, Zhang Z, Zhao X, et al. GVCNN: Group-view convolutional neural networks for 3D shape recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 264-272.

• https://github.com/jongchyisu/mvcnn_pytorch

• https://github.com/ace19-dev/gvcnn

• https://github.com/Cadene/pretrained-models.pytorch