A Pytorch implementation of Multi-view Dual Attention Network for 3D Object Recognitionn (MVDAN)
In this paper, the 3D object recognition problem is converted to multi-view 2D image classification problem. For each 3D object, there are multiple images taken from different views
- Python 3.6
- PyTorch 1.2.0
- numpy
- ModelNet CAD data can be found at Princeton
- ModelNet40 12-view png images can be downloaded at modelnet40_images_new_12x (1.6GB)
- You can also create 3-view png images and 6-view png images by reducing the number of 12 views
python train.py -name MVDAN -num_models 1000 -weight_decay 0.0001 -num_views 12 -cnn_name resnet50