MapUncertaintyPrediction

This repository contains the official implementation of Producing and Leveraging Online Map Uncertainty in Trajectory Prediction published in CVPR 2024 [Award Candidate].

Getting Started

• Environment Setup
• Prepare Dataset
• Mapping Train and Eval
• Merge Map and Trajectory Dataset
• Trajectory Train and Eval
• Visualization

Results

Mapping checkpoints are here. Trajectory prediction checkpoints are here.

Dataset

I have uploaded all datasets (complete) for MapTR, StreamMapNet, MapTRv2 and MapTRv2 CL. They are around 500GB each (StreamMapNet is around 200GB). They are large because they also contain the intermediate BEV features. After downloading, you can simply remove the BEV features to save space. You can download them through Hugging Face. They are located at here which should be easier to download.

Dataset Structure is as follows:

mapbevprediction
├── stream_bev/
├── maptr_bev/
├── maptrv2_bev/
│   ├── mini_val/
│   |   ├── data/
│   |   |   ├── scene-{scene_id}.pkl
│   ├── train/
│   ├── val/
├── maptrv2_cent_bev/

News

• Mar 10, 2025: Another extention of this work that I am involved in is available here. Check it out!
• July 01, 2024: An extention of this work has been published at ECCV 2024. Check it out!
• May 29, 2024: This work has been slected as a Best Paper Award Finalist at CVPR 2024! Top 0.2% of all submissions (24/11,532)

Catalog

• Visualization Code
• Code release
  • MapTR
  • MapTRv2
  • StreamMapNet
  • HiVT
  • DenseTNT
• Initialization

Citation

If you found this repository useful, please consider citing our work:

@Inproceedings{GuSongEtAl2024,
  author    = {Gu, Xunjiang and Song, Guanyu and Gilitschenski, Igor and Pavone, Marco and Ivanovic, Boris},
  title     = {Producing and Leveraging Online Map Uncertainty in Trajectory Prediction},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2024}
}

License

This repository is licensed under Apache 2.0.