slamcore / ActiveStereoNet

Active stereo net implementation for active tartanair and D435i datsets

ActiveStereoNet

This repository builds upon the open source pytorch Active stereo net implementation and extends it to active tartanair and D435i datsets.

Paper

ActiveStereoNet: End-to-End Self-Supervised Learning for Active Stereo Systems

Requirments

CUDA = v11.1

CuDNN >= v8.2.1

Python > 3.8

Pytorch

Torchvision

Dataset

Datasets used:

1. D435i dataset (real data)
2. Active TartanAir dataset (virtual data)

Please, use the links provided to download the datasets and update the data_root field in the Options/*.json files.

Usage

To train on D435i sequences:

sh d435i.sh

To train on Active Tartanair sequences:

sh tartanair.sh