PyTorch code for MVCNN

Code is tested on Python 3.6 and PyTorch 0.4.1

First, download images and put it under modelnet40_images_new_12x:
Shaded Images (1.6GB)

Command for training:
python train_mvcnn.py -name mvcnn -num_models 1000 -weight_decay 0.001 -num_views 12 -cnn_name vgg11

Project webpage
Depth Images (1.6GB)

Blender script for rendering shaded images
Blender script for rendering depth images

Reference

A Deeper Look at 3D Shape Classifiers
Jong-Chyi Su, Matheus Gadelha, Rui Wang, and Subhransu Maji
Second Workshop on 3D Reconstruction Meets Semantics, ECCV, 2018

Multi-view Convolutional Neural Networks for 3D Shape Recognition
Hang Su, Subhransu Maji, Evangelos Kalogerakis, and Erik Learned-Miller,
International Conference on Computer Vision, ICCV, 2015