This repo contains code for training and evaluating OmniSyn.
OmniSyn predicts depth from two panoramas and then renders meshes using PyTorch3D from intermediate positions.
- Install PyTorch and PyTorch3d
- Download our processed carla dataset.
Currently, the main scripts are:
train_depth.py
which trains the depth estimator only.train_inpainting.py
to train the inpainting component.
- Install v0.2.5 of PyTorch3d
pip install 'git+https://github.com/facebookresearch/pytorch3d.git@v0.2.5'