CoCoNets: Continuous Contrastive 3D Scene Representations

** (Code in progress) ** This repository contains the code that accompanies our CVPR 2021 paper CoCoNets: Continuous Contrastive 3D Scene Representations

Installation

First you have to make sure that you have all dependencies in place. The simplest way to do so, is by running the following command:

pip install -r requirements.txt

For our implmentation we use torch==1.12.1+cu113 and torchvision==0.13.1+cu113, although feel free to use your favorite version.

Dataset and checkpoints

You can find the CARLA and Kitti dataset, along with the pretrained checkpoints here.

For instance download, the CARLA dataset zip file from here

Further extract the dataset zip and then edit the dataset_location variable here

Multiview Training

Checkpoints will be saved in pytorch_disco/checkpoints directory. Any checkpoint model that you want to be loaded should be placed at the end in pretrained_nets_carla.py file.

python main.py cm --en trainer_implicit --rn <run_name>

Evaluation on Tracking task

python main.py cz --en tester_pointfeat_dense --rn <run_name>