HariyaNobuki / CoverNet

Implementation of multi-modal path prediction - CoverNet with Pytorch and nuscenes dataset (Not completed)

CoverNet Implementation (WARN : not completed)

This repository contains an implementation of CoverNet and Nuscenes dataset processing.

Phan-Minh, Tung, et al. "Covernet: Multimodal behavior prediction using trajectory sets." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020.

Setup

1. Download Repository git clone https://github.com/hyerim-mmc/CoverNet.git

2. Download Nuscenes Dataset_Full dataset(v1.0)

   • Dataset Architecture should be as follows

   ${project_folder_name}
     |__data
       |__sets
         |__nuscenes
            |__maps
            |__samples
            |__sweeps
            |__v1.0-mini
            |__detection.json
   

3. Download Map expansion

   • Extract the contents (folders basemap, expansion and prediction) to your nuScenes maps folder.

4. Download Nuscenes-devkit

5. Download Pytorch

6. Download Tensorboard pip install tensorboard or conda install tensorboard

Run

1. Download the trajectory sets
2. Write own parsing/learning configuration covernet_config.json
3. Run python train.py
   • Results will be saved in result folder
   • Check Tensorboard results using tensorboard --logdir=./result/tensorboard and click the website