DeepV2D (PyTorch)

This is PyTorch implementation of the paper "DeepV2D: Video to Depth with Differentiable Structure from Motion"(ICLR 2020).

Original tensorflow project is at here.

Installation

Develop environment:

OS: Ubuntu 20.04.4
CUDA: 11.4

Install dependencies:

pip install -r requirements.txt

Install LieTorch for SE3 operations to pose operation:

cd lietorch
python setup.py install
./run_tests.sh # for test