Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose

Release Notes

The official PyTorch implementation of Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose, publish on NeuralIPS 2021.

Installation

To install required libs:

git clone https://github.com/Angtian/NeuralVS.git
cd NeuralVS
pip install -r requirements.txt

We use the same data preprocess as NeMo does. You can run the data preprocess part in NeMo repo or run the following code:

git clone https://github.com/Angtian/NeMo.git
cd NeMo
chmod +x PrepareData.sh
./PrepareData.sh
cd ..
mv ./NeMo/data ./

Matching using Single Image

Here we provide the code to run the pose matching experiment using single anchor image (section 4.3). To run the experiment using ImageNet pretrained backbone:

python .\code\SingleAnchorMatching.py --do_plot

To run the experiment with other backbone:

python .\code\SingleAnchorMatching.py --load_path {Path_to_saved_model} --do_plot

Code for Semi-supervised learning

Coming Soon