Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

https://github.com/hustvl/MapTR

[ICLR'23 Spotlight] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction
https://github.com/hustvl/MapTR

autonomous-driving bev end-to-end iclr2023 online-hdmap-construction real-time shape-representation transformer vectorized-hdmap

Last synced: about 2 months ago
JSON representation

[ICLR'23 Spotlight] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction

Host: GitHub
URL: https://github.com/hustvl/MapTR
Owner: hustvl
License: mit
Created: 2022-07-28T02:20:43.000Z (almost 2 years ago)
Default Branch: main
Last Pushed: 2023-10-19T06:02:43.000Z (8 months ago)
Last Synced: 2024-04-28T06:05:24.893Z (2 months ago)
Topics: autonomous-driving, bev, end-to-end, iclr2023, online-hdmap-construction, real-time, shape-representation, transformer, vectorized-hdmap
Language: Python
Homepage:
Size: 8.95 MB
Stars: 913
Watchers: 40
Forks: 139
Open Issues: 106
Metadata Files:
- Readme: README.md
- License: LICENSE

Lists

Awesome-Self-Driving - MapTR - [ICLR'23 Spotlight] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction (6. Detection / 3.4. Others)
awesome-stars - hustvl/MapTR - [ICLR'23 Spotlight] MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction (Python)
awesome-hd-map-construction - [code

README

        


MapTR 



An End-to-End Framework for Online Vectorized HD Map Construction


[Bencheng Liao](https://github.com/LegendBC)^1,2,3 \*, [Shaoyu Chen](https://scholar.google.com/citations?user=PIeNN2gAAAAJ&hl=en&oi=sra)^1,3 \*, [Yunchi Zhang](https://github.com/zyc10ud)^1,3 , [Bo Jiang](https://github.com/rb93dett)^1,3 ,[Tianheng Cheng](https://scholar.google.com/citations?user=PH8rJHYAAAAJ&hl=zh-CN)^1,3, [Qian Zhang](https://scholar.google.com/citations?user=pCY-bikAAAAJ&hl=zh-CN)³, [Wenyu Liu](http://eic.hust.edu.cn/professor/liuwenyu/)¹, [Chang Huang](https://scholar.google.com/citations?user=IyyEKyIAAAAJ&hl=zh-CN)³, [Xinggang Wang](https://xwcv.github.io)^{1 :email:}

 

¹ School of EIC, HUST, ² Institute of Artificial Intelligence, HUST, ³ Horizon Robotics

(\*) equal contribution, (^:email:) corresponding author.

ArXiv Preprint ([arXiv 2208.14437](https://arxiv.org/abs/2208.14437))

[openreview ICLR'23](https://openreview.net/forum?id=k7p_YAO7yE), accepted as **ICLR Spotlight**

extended ArXiv Preprint MapTRv2 ([arXiv 2308.05736](https://arxiv.org/abs/2308.05736))



#

### News

* **`Aug. 31th, 2023`:** initial MapTRv2 is released at ***maptrv2*** branch. Please run `git checkout maptrv2` to use it.

* **`Aug. 14th, 2023`:** As required by many researchers, the code of MapTR-based map annotation framework (VMA) will be released at https://github.com/hustvl/VMA recently.

* **`Aug. 10th, 2023`:** We release [MapTRv2](https://arxiv.org/abs/2308.05736) on Arxiv. MapTRv2 demonstrates much stronger performance and much faster convergence. To better meet the requirement of the downstream planner (like [PDM](https://github.com/autonomousvision/nuplan_garage)), we introduce an extra semantic——centerline (using path-wise modeling proposed by [LaneGAP](https://github.com/hustvl/LaneGAP)). Code & model will be released in late August. Please stay tuned!

* **`May. 12th, 2023`:** MapTR now support various bevencoder, such as [BEVFormer encoder](projects/configs/maptr/maptr_tiny_r50_24e_bevformer.py) and [BEVFusion bevpool](projects\configs\maptr\maptr_tiny_r50_24e_bevpool.py). Check it out!

* **`Apr. 20th, 2023`:** Extending MapTR to a general map annotation framework ([paper](https://arxiv.org/pdf/2304.09807.pdf), [code](https://github.com/hustvl/VMA)), with high flexibility in terms of spatial scale and element type.

* **`Mar. 22nd, 2023`:** By leveraging MapTR, VAD ([paper](https://arxiv.org/abs/2303.12077), [code](https://github.com/hustvl/VAD))  models the driving scene as fully vectorized representation, achieving SoTA end-to-end planning performance!

* **`Jan. 21st, 2023`:** MapTR is accepted to ICLR 2023 as **Spotlight Presentation**!

* **`Nov. 11st, 2022`:** We release an initial version of MapTR.

* **`Aug. 31st, 2022`:** We released our paper on Arxiv. Code/Models are coming soon. Please stay tuned! ☕️

## Introduction

MapTR/MapTRv2 is a simple, fast and strong online vectorized HD map construction framework.


![framework](assets/teaser.png "framework")

High-definition (HD) map provides abundant and precise static environmental information of the driving scene, serving as a fundamental and indispensable component for planning in autonomous driving system. In this paper, we present **Map** **TR**ansformer, an end-to-end framework for online vectorized HD map construction. We propose a unified permutation-equivalent modeling approach, i.e., modeling map element as a point set with a group of equivalent permutations, which accurately describes the shape of map element and stabilizes the learning process. We design a hierarchical query embedding scheme to flexibly encode structured map information and perform hierarchical bipartite matching for map element learning. To speed up convergence, we further introduce auxiliary one-to-many matching and dense supervision. The proposed method well copes with various map elements with arbitrary shapes. It runs at real-time inference speed and achieves state-of-the-art performance on both nuScenes and Argoverse2 datasets. Abundant qualitative results show stable and robust map construction quality in complex and various driving scenes.

## Models

> Results from the [MapTRv2 paper](https://arxiv.org/abs/2308.05736)

![comparison](assets/comparison.png "comparison")

| Method | Backbone | Lr Schd | mAP| FPS|

| :---: | :---: | :---: | :---: | :---: 

| MapTR | R18 | 110ep | 45.9 | 35.0| 

| MapTR | R50 | 24ep | 50.3 | 15.1| 

| MapTR | R50 | 110ep | 58.7|15.1|

| MapTRv2 | R18 | 110ep | 52.3 | 33.7|

| MapTRv2 | R50 | 24ep | 61.5 | 14.1|

| MapTRv2 | R50 | 110ep | 68.7 | 14.1|

| MapTRv2 | V2-99 | 110ep | 73.4 | 9.9|

**Notes**: 

- FPS is measured on NVIDIA RTX3090 GPU with batch size of 1 (containing 6 view images).

- All the experiments are performed on 8 NVIDIA GeForce RTX 3090 GPUs. 

> Results from this repo. 

### MapTR

 nuScenes dataset


| Method | Backbone | BEVEncoder |Lr Schd | mAP| FPS|memory | Config | Download |

| :---: | :---: | :---: | :---: |  :---: | :---:|:---:| :---: | :---: |

| MapTR-nano | R18 |GKT | 110ep |46.3  |35.0| 11907M (bs 24) |[config](projects/configs/maptr/maptr_nano_r18_110e.py) |[model](https://drive.google.com/file/d/1-wVO1pZhFif2igJoz-s451swQvPSto2m/view?usp=sharing) / [log](https://drive.google.com/file/d/1Hd25seDQKn8Vv6AQxPfSoiu-tY2i4Haa/view?usp=sharing) |

| MapTR-tiny | R50 | GKT |24ep | 50.0 |15.1| 10287M (bs 4) | [config](projects/configs/maptr/maptr_tiny_r50_24e.py)|[model](https://drive.google.com/file/d/1n1FUFnRqdskvmpLdnsuX_VK6pET19h95/view?usp=share_link) / [log](https://drive.google.com/file/d/1nvPkk0EMHV8Q82E9usEKKYx7P38bCx1U/view?usp=share_link) |

| MapTR-tiny | R50 |GKT | 110ep | 59.3 |15.1| 10287M (bs 4)|[config](projects/configs/maptr/maptr_tiny_r50_110e.py) |[model](https://drive.google.com/file/d/1SCF93LEEmXU0hMwPiUz9p2CWbL1FpB1h/view?usp=share_link) / [log](https://drive.google.com/file/d/1TQ4j_0Sf2ipzeYsEZZAHYzX4dCUaBqyp/view?usp=share_link) |

| MapTR-tiny | Camera & LiDAR | GKT |24ep | 62.7 | 6.0 | 11858M (bs 4)|[config](projects/configs/maptr/maptr_tiny_fusion_24e.py) |[model](https://drive.google.com/file/d/1CFlJrl3ZDj3gIOysf5Cli9bX5LEYSYO4/view?usp=share_link) / [log](https://drive.google.com/file/d/1rb3S4oluxdZjNm2aJ5lBH23jrkYIaJbC/view?usp=share_link) |

| MapTR-tiny | R50 | bevpool |24ep | 50.1 | 14.7 | 9817M (bs 4)|[config](projects/configs/maptr/maptr_tiny_r50_24e_bevpool.py) |[model](https://drive.google.com/file/d/16PK9XohV55_3qPVDtpXIl4_Iumw9EnfA/view?usp=sharing) / [log](https://drive.google.com/file/d/14nioV3_VV9KehmxK7XcAHxM8X6JH5WIr/view?usp=sharing) |

| MapTR-tiny | R50 | bevformer |24ep | 48.7 | 15.0 | 10219M (bs 4)|[config](projects/configs/maptr/maptr_tiny_r50_24e_bevformer.py) |[model](https://drive.google.com/file/d/1y-UBwGBSb2xiV40AuQEBhB-xJyV7VusX/view?usp=sharing) / [log](https://drive.google.com/file/d/1r35bRhTGVtyZTP8drXBTOIhLYGCzjEaF/view?usp=sharing) |

| MapTR-tiny⁺ | R50 | GKT |24ep | 51.3 | 15.1 | 15158M (bs 4)|[config](projects/configs/maptr/maptr_tiny_r50_24e_t4.py) |[model](https://drive.google.com/file/d/1SWmBriDG8vwLXmWTHGVdrRUrDBxzGa3a/view?usp=drive_link) / [log](https://drive.google.com/file/d/1pJmNL7AhmkwA5Er6nZVpw7qApEhcwMFY/view?usp=drive_link) |

| MapTR-tiny⁺ | R50 | bevformer |24ep | 53.3 | 15.0 | 15087M (bs 4)|[config](projects/configs/maptr/maptr_tiny_r50_24e_bevformer_t4.py) |[model](https://drive.google.com/file/d/1sbXTawEbpV61TwVULCMRRDTLYzZ6SL7U/view?usp=sharing) / [log](https://drive.google.com/file/d/1YGI_X6Cb2zV13CHMsDvEs8RJRMzeiUzM/view?usp=sharing) |

**Notes**: 

-  ⁺ means that we introduce temporal setting.

### MapTRv2

Please `git checkout maptrv2` and follow the install instruction to use following checkpoint

 nuScenes dataset


| Method | Backbone | BEVEncoder |Lr Schd | mAP| FPS|memory | Config | Download |

| :---: | :---: | :---: | :---: |  :---: | :---:|:---:| :---: | :---: |

| MapTRv2| R50 |bevpool | 24ep | 61.4 |14.1| 19426M (bs 24) |[config](https://github.com/hustvl/MapTR/blob/maptrv2/projects/configs/maptrv2/maptrv2_nusc_r50_24ep.py) |[model](https://drive.google.com/file/d/1AmQ3fT-J-MM4B8kh_9Gm2G5guM92Agww/view?usp=sharing) / [log](https://drive.google.com/file/d/1rrAXza6FTYUs8kfr5126qWU6-FNGGMwD/view?usp=sharing) |

| MapTRv2*| R50 |bevpool | 24ep | 54.3 |WIP| 20363M (bs 24) |[config](https://github.com/hustvl/MapTR/blob/maptrv2/projects/configs/maptrv2/maptrv2_nusc_r50_24ep_w_centerline.py) |[model](https://drive.google.com/file/d/1m02OKAKPhzMOaSu_4STVcepY8jbE7v3o/view?usp=sharing) / [log](https://drive.google.com/file/d/1cEV7sfiWS0-9Uu1eQEt2xm77l4mAuHMM/view?usp=sharing) |

 Argoverse2 dataset


| Method | Backbone | BEVEncoder |Lr Schd | mAP| FPS|memory | Config | Download |

| :---: | :---: | :---: | :---: |  :---: | :---:|:---:| :---: | :---: |

| MapTRv2| R50 |bevpool | 6ep | 64.3 |14.1| 20580 (bs 24) |[config](https://github.com/hustvl/MapTR/blob/maptrv2/projects/configs/maptrv2/maptrv2_av2_3d_r50_6ep.py) |[model](https://drive.google.com/file/d/18-uyyP4ijjMRizSSOsV0GnPgtMNlPfG5/view?usp=sharing) / [log](https://drive.google.com/file/d/1Z5-4ATksKZbcfnGLnEc5aEsxA79GlqRN/view?usp=sharing) |

| MapTRv2*| R50 |bevpool | 6ep | 61.3 |WIP| 21515 (bs 24) |[config](https://github.com/hustvl/MapTR/blob/maptrv2/projects/configs/maptrv2/maptrv2_av2_3d_r50_6ep_w_centerline.py) |[model](https://drive.google.com/file/d/1wXugPxU8HKeGxPAyFdgb53D5_zGmiCen/view?usp=sharing) / [log](https://drive.google.com/file/d/1vm60KzlGrbz5IEAXgKyqxrOydUF1F-6E/view?usp=sharing) |

**Notes**: 

- \* means that we introduce an extra semantic——centerline (using path-wise modeling proposed by [LaneGAP](https://github.com/hustvl/LaneGAP)).

## Qualitative results on nuScenes val split and Argoverse2 val split

 MapTR/MapTRv2 maintains stable and robust map construction quality in various driving scenes.


![visualization](assets/MapTRv2_av2_visualizations.png "visualization")

### *MapTRv2 on whole nuScenes val split*

[**Youtube**](https://www.youtube.com/watch?v=s7McToPNlJ4)

### *MapTRv2 on whole Argoverse2 val split*

[**Youtube**](https://www.youtube.com/watch?v=nC8W_2BZuys)

### *End-to-end Planning based on MapTR*

https://user-images.githubusercontent.com/26790424/229679664-0e9ba5e8-bf2c-45e0-abbc-36d840ee5cc9.mp4

## Getting Started

- [Installation](docs/install.md)

- [Prepare Dataset](docs/prepare_dataset.md)

- [Train and Eval](docs/train_eval.md)

- [Visualization](docs/visualization.md)

## Catalog

- [x] temporal modules

- [x] centerline detection & topology support (refer to ***maptrv2*** branch)

- [x] multi-modal checkpoints

- [x] multi-modal code

- [ ] lidar modality code

- [x] argoverse2 dataset 

- [x] Nuscenes dataset 

- [x] MapTR checkpoints

- [x] MapTR code

- [x] Initialization

## Acknowledgements

MapTR is based on [mmdetection3d](https://github.com/open-mmlab/mmdetection3d). It is also greatly inspired by the following outstanding contributions to the open-source community: [BEVFusion](https://github.com/mit-han-lab/bevfusion), [BEVFormer](https://github.com/fundamentalvision/BEVFormer), [HDMapNet](https://github.com/Tsinghua-MARS-Lab/HDMapNet), [GKT](https://github.com/hustvl/GKT), [VectorMapNet](https://github.com/Mrmoore98/VectorMapNet_code).

## Citation

If you find MapTR is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

```bibtex

@inproceedings{MapTR,

  title={MapTR: Structured Modeling and Learning for Online Vectorized HD Map Construction},

  author={Liao, Bencheng and Chen, Shaoyu and Wang, Xinggang and Cheng, Tianheng, and Zhang, Qian and Liu, Wenyu and Huang, Chang},

  booktitle={International Conference on Learning Representations},

  year={2023}

}

```

```bibtex

@article{maptrv2,

  title={MapTRv2: An End-to-End Framework for Online Vectorized HD Map Construction},

  author={Liao, Bencheng and Chen, Shaoyu and Zhang, Yunchi and Jiang, Bo and Zhang, Qian and Liu, Wenyu and Huang, Chang and Wang, Xinggang},

  journal={arXiv preprint arXiv:2308.05736},

  year={2023}

}

```

```bibtex

 @article{lanegap,

  title={Lane Graph as Path: Continuity-preserving Path-wise Modeling for Online Lane Graph Construction},

  author={Bencheng Liao and Shaoyu Chen and Bo Jiang and Tianheng Cheng and Qian Zhang and Wenyu Liu and Chang Huang and Xinggang Wang},

  journal={arXiv preprint arXiv:2303.08815},

  year={2023}

}

```