You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
-:oncoming_automobile:**Planning-oriented philosophy**: UniAD is a Unified Autonomous Driving algorithm framework devised following a planning-oriented philosophy. Instead of standalone modular design and multi-task learning, perception, prediciton and planning tasks/components should opt in and be prioritized hierarchically, and we demonstrate the performance can be enhanced to a new level.
39
-
-:trophy:**SOTA performance**: All tasks among UniAD achieve SOTA performance, especially prediction and planning (motion: 0.71m minADE, occ: 63.4% IoU-n., plan: 0.31% avg.Col)
45
+
## Highlights <aname="high"></a>
40
46
41
-
## News
47
+
-:oncoming_automobile:**Planning-oriented philosophy**: UniAD is a Unified Autonomous Driving algorithm framework following a planning-oriented philosophy. Instead of standalone modular design and multi-task learning, we cast a series of tasks, including perception, prediction and planning tasks hierarchically.
48
+
-:trophy:**SOTA performance**: All tasks within UniAD achieve SOTA performance, especially prediction and planning (motion: 0.71m minADE, occ: 63.4% IoU, planning: 0.31% avg.Col)
42
49
43
-
- Code & model release: We are actively re-organizing the codebase for better readability. The estimated time is late March. Please stay tuned!
44
-
- About the title: To avoid confusion about the "goal", we change the title from "Goal-oriented" to "Planning-oriented" as suggested by the reviewers. We originally use "goal" to indicate the final safe planning in an AD pipeline, rather than "goal-point" -- the destination of a sequence of actions.
45
-
-[2023/03/21]:rocket::rocket: UniAD paper is accepted by CVPR 2023, as an **award candidate** (12 out of 9155 submissions and 2360 accepted papers)!
46
-
-[2022/12/21] UniAD [paper](https://arxiv.org/abs/2212.10156) is available on arXiv!
50
+
## News <aname="news"></a>
47
51
48
-
<!--
49
-
## Getting started
52
+
-**`Paper Title Change`**: To avoid confusion with the "goal-point" navigation in Robotics, we change the title from "Goal-oriented" to "Planning-oriented" suggested by Reviewers. Thank you!
53
+
-[2023/04]**_Estimated_**. Model checkpoints release `v2.0`
54
+
55
+
56
+
-[2023/03/29] Code & model initial release `v1.0`
57
+
-[2023/03/21]:rocket::rocket: UniAD is accepted by CVPR 2023, as an **Award Candidate** (12 out of 2360 accepted papers)!
58
+
-[2022/12/21] UniAD [paper](https://arxiv.org/abs/2212.10156) is available on arXiv.
## Results and Pre-trained Models <aname="models"></a>
82
+
UniAD is trained in two stages. Pretrained checkpoints of both stages will be released and the results of each model are listed in the following tables.
83
+
84
+
### Stage-one: Perception training
85
+
> We first train the perception modules (i.e., track and map) to obtain a stable initlization for the next stage.
Pre-trained models and results under main metrics are provided below. We refer you to the [paper](https://arxiv.org/abs/2212.10156) for more details.
94
+
95
+
### Stage-two: End-to-end training
96
+
> We optimize all task modules together, including track, map, motion, occupancy and planning.
97
+
98
+
<!--
99
+
Pre-trained models and results under main metrics are provided below. We refer you to the [paper](https://arxiv.org/abs/2212.10156) for more details. -->
* Download the checkpoints you need into `UniAD/ckpts/` directory.
109
+
* You can evaluate these checkpoints to reproduce the results, following the `evaluation` section in [TRAIN_EVAL.md](docs/TRAIN_EVAL.md).
110
+
* You can also initialize your own model with the provided weights. Change the `load_from` field to `path/of/ckpt` in the config and follow the `train` section in [TRAIN_EVAL.md](docs/TRAIN_EVAL.md) to start training.
111
+
112
+
113
+
### Model Structure
114
+
The overall pipeline of UniAD is controlled by [uniad_e2e.py](projects/mmdet3d_plugin/uniad/detectors/uniad_e2e.py) which coordinates all the task modules in `UniAD/projects/mmdet3d_plugin/uniad/dense_heads`. If you are interested in the implementation of a specific task module, please refer to its corresponding file, e.g., [motion_head](projects/mmdet3d_plugin/uniad/dense_heads/motion_head.py).
-[ ] Separating BEV encoder and tracking module [Est. 2023/04]
119
+
-[ ] Support larger batch size [Est. 2023/04]
120
+
-[ ] (Long-term) Improve flexibility for future extensions
121
+
-[ ] All configs & checkpoints
122
+
-[x] Code initialization
67
123
68
-
All assets and code are under the [Apache 2.0 license](https://github.com/OpenDriveLab/UniAD/blob/master/LICENSE) unless specified otherwise.
69
124
70
-
## Citation
125
+
## License <aname="license"></a>
126
+
127
+
All assets and code are under the [Apache 2.0 license](./LICENSE) unless specified otherwise.
128
+
129
+
## Citation <aname="citation"></a>
71
130
72
131
Please consider citing our paper if the project helps your research with the following BibTex:
73
132
74
133
```bibtex
75
-
@inproceedings{uniad,
134
+
@inproceedings{hu2023_uniad,
76
135
title={Planning-oriented Autonomous Driving},
77
136
author={Yihan Hu and Jiazhi Yang and Li Chen and Keyu Li and Chonghao Sima and Xizhou Zhu and Siqi Chai and Senyao Du and Tianwei Lin and Wenhai Wang and Lewei Lu and Xiaosong Jia and Qiang Liu and Jifeng Dai and Yu Qiao and Hongyang Li},
78
137
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
@@ -82,10 +141,9 @@ Please consider citing our paper if the project helps your research with the fol
Download nuScenes V1.0 full dataset data, CAN bus and map(v1.3) extensions [HERE](https://www.nuscenes.org/download), following the steps below to prepare the data.
5
+
6
+
7
+
**Download nuScenes, CAN_bus and Map extensions**
8
+
```shell
9
+
cd UniAD
10
+
mkdir data
11
+
# Download nuScenes V1.0 full dataset data directly to (or soft link to) UniAD/data/
12
+
# Download CAN_bus and Map(v1.3) extensions directly to (or soft link to) UniAD/data/nuscenes/
13
+
```
14
+
15
+
**Prepare UniAD data infos**
16
+
17
+
*Option1: We have already prepared the off-the-shelf data infos for you:*
0 commit comments