SICGAN

Single Image 3D Reconstruction based on Conditional GAN [report]

Code for EECS 504: Introduction to Computer Vision Project

Team Members:
Danish Syed^*, Hansal Shah^* Manas Buragohain^*,
* Indicates equal contribution

Requirements

• PyTorch 1.4.0
• PyTorch 3D
• CUDA 10.1
• cuDNN 7.6

Training

Pixel2Mesh Baseline
python train_p2m.py --config-yml config/train_p2m.yml

Vanilla SICGAN
python train_p2m_gan.py --config-yml config/train_p2m_gan.yml

SICGAN with random noise
python train_p2m_randgan.py --config-yml config/train_p2m_randgan.yml

License

The SICGAN is realeased under MIT License