Skip to content

Commit f3ae74d

Browse files
yanglei18yanglei18
committed
update nuscense_bevheight_plus
1 parent d3dc93f commit f3ae74d

9 files changed

+976
-493
lines changed

‎README.md

+134-7
Original file line numberDiff line numberDiff line change
@@ -44,9 +44,7 @@
4444
# News
4545

4646
- [2024/09/06] Both arXiv and codebase are released!
47-
# Incoming
4847

49-
- [ ] Release the pretrained models
5048

5149
<br>
5250

@@ -81,14 +79,139 @@ Eval `BEVDepth / BEVHeight / BEVHeight++` with 8 GPUs
8179
```
8280
python [EXP_PATH] --ckpt_path [CKPT_PATH] -e -b 8 --gpus 8
8381
```
84-
**For more specific training and evaluation commands, please refer to the `train_scripts` directory.**
82+
>* **For more specific training and evaluation commands, please refer to the `train_scripts` directory.**
83+
84+
>* **For experiments on nuScenes dataset, please refer to the `nuscense_bevheight_plus` directory and this [document](nuscense_bevheight_plus/README.md)
85+
.**
8586

8687
# Experimental Results
8788
- **DAIR-V2X-I Dataset**
88-
89+
<div align=left>
90+
<table>
91+
<tr align=center>
92+
<td rowspan="3">Method</td>
93+
<td rowspan="3" align=center>Config</td>
94+
<td colspan="3" align=center>Car</td>
95+
<td colspan="3" align=center>Pedestrain</td>
96+
<td colspan="3" align=center>Cyclist</td>
97+
</tr>
98+
<tr align=center>
99+
<td colspan="3" align=center>[email protected]</td>
100+
<td colspan="3" align=center>[email protected]</td>
101+
<td colspan="3" align=center>[email protected]</td>
102+
</tr>
103+
<tr align=center>
104+
<td>Easy</td>
105+
<td>Mod.</td>
106+
<td>Hard</td>
107+
<td>Easy</td>
108+
<td>Mod.</td>
109+
<td>Hard</td>
110+
<td>Easy</td>
111+
<td>Mod.</td>
112+
<td>Hard</td>
113+
</tr>
114+
<tr align=center>
115+
<td rowspan="1">BEVDepth</td>
116+
<td><a href=exps/dair-v2x/bev_depth_lss_r101_864_1536_256x256_102.py>config</td>
117+
<td>75.50</td>
118+
<td>63.58</td>
119+
<td>63.67</td>
120+
<td>34.95</td>
121+
<td>33.42</td>
122+
<td>33.27</td>
123+
<td>55.67</td>
124+
<td>55.47</td>
125+
<td>55.34</td>
126+
</tr>
127+
<tr align=center>
128+
<td rowspan="1">BEVHeight</td>
129+
<td><a href=exps/dair-v2x/bev_height_lss_r101_864_1536_256x256_102.py>config</td>
130+
<td>77.48</td>
131+
<td>65.46</td>
132+
<td>65.53</td>
133+
<td>41.22</td>
134+
<td>39.29</td>
135+
<td>39.46</td>
136+
<td>60.23</td>
137+
<td>60.08</td>
138+
<td>60.54</td>
139+
</tr>
140+
<tr align=center>
141+
<td rowspan="1">BEVHeight++</td>
142+
<td><a href=exps/dair-v2x/bev_height_plus_lss_r101_864_1536_256x256_102.py>config</td>
143+
<td><b>79.31</b></td>
144+
<td><b>68.62</b></td>
145+
<td><b>68.68</b></td>
146+
<td><b>42.87</b></td>
147+
<td><b>40.88</b></td>
148+
<td><b>41.06</b></td>
149+
<td><b>60.76</b></td>
150+
<td><b>60.52</b></td>
151+
<td><b>61.01</b></td>
152+
</tr>
153+
<table>
154+
</div>
89155

90156
- **Rope3D Dataset**
91-
157+
<div align=center>
158+
<table>
159+
<tr align=center>
160+
<td rowspan="2">Method</td>
161+
<td rowspan="2" align=center>Config</td>
162+
<td colspan="2" align=center>Car | [email protected]</td>
163+
<td colspan="2" align=center>Big Vehicle | [email protected]</td>
164+
<td colspan="2" align=center>Car | [email protected]</td>
165+
<td colspan="2" align=center>Big Vehicle | [email protected]</td>
166+
</tr>
167+
<tr align=center>
168+
<td>AP</td>
169+
<td>Rope</td>
170+
<td>AP</td>
171+
<td>Rope</td>
172+
<td>AP</td>
173+
<td>Rope</td>
174+
<td>AP</td>
175+
<td>Rope</td>
176+
</tr>
177+
<tr align=center>
178+
<td rowspan="1">BEVDepth</td>
179+
<td><a href=exps/rope3d/bev_depth_lss_r101_864_1536_256x256_102.py>config</td>
180+
<td>69.63</td>
181+
<td>74.70</td>
182+
<td>45.02</td>
183+
<td>54.64</td>
184+
<td>42.56</td>
185+
<td>53.05</td>
186+
<td>21.47</td>
187+
<td>35.82</td>
188+
</tr>
189+
<tr align=center>
190+
<td rowspan="1">BEVHeight</td>
191+
<td><a href=exps/rope3d/bev_height_lss_r101_864_1536_256x256_102.py>config</td>
192+
<td>74.60</td>
193+
<td>78.72</td>
194+
<td>48.93</td>
195+
<td>57.70</td>
196+
<td>45.73</td>
197+
<td>55.62</td>
198+
<td>23.07</td>
199+
<td>37.04</td>
200+
</tr>
201+
<tr align=center>
202+
<td rowspan="1">BEVHeight++</td>
203+
<td><a href=exps/rope3d/bev_height_plus_lss_r101_864_1536_256x256_102.py>config</td>
204+
<td><b>76.12</b></td>
205+
<td><b>80.91</b></td>
206+
<td><b>50.11</b></td>
207+
<td><b>59.92</b></td>
208+
<td><b>47.03</b></td>
209+
<td><b>57.77</b></td>
210+
<td><b>24.43</b></td>
211+
<td><b>39.57</b></td>
212+
</tr>
213+
</table>
214+
</div>
92215

93216
- **KITTI Dataset**
94217
<div align=left>
@@ -97,7 +220,7 @@ python [EXP_PATH] --ckpt_path [CKPT_PATH] -e -b 8 --gpus 8
97220
<td rowspan="3">Method</td>
98221
<td colspan="3" align=center>AP|3D</td>
99222
<td colspan="3" align=center>AP|BEV</td>
100-
<td rowspan="3" align=center>config</td>
223+
<td rowspan="3" align=center>Config</td>
101224
<td rowspan="3" align=center>model ckpt</td>
102225
</tr>
103226
<tr align=center>
@@ -378,10 +501,14 @@ This project is not possible without the following codebases.
378501
* [BEVHeight](https://github.com/ADLab-AutoDrive/BEVHeight)
379502
* [BEVDepth](https://github.com/Megvii-BaseDetection/BEVDepth)
380503
* [DAIR-V2X](https://github.com/AIR-THU/DAIR-V2X)
504+
* [BEVDet](https://github.com/HuangJunJie2017/BEVDet)
505+
* [mmdetection3d](https://github.com/open-mmlab/mmdetection3d)
381506
* [pypcd](https://github.com/dimatura/pypcd)
507+
* [PanopticBEV](https://github.com/abhi1kumar/SeaBird/tree/main/PanopticBEV)
508+
382509

383510
# Citation
384-
If you use BEVHeight in your research, please cite our work by using the following BibTeX entry:
511+
If you use BEVHeight++ in your research, please cite our work by using the following BibTeX entry:
385512
```
386513
@article{yang2023bevheight++,
387514
title={Bevheight++: Toward robust visual centric 3d object detection},

‎docs/prepare_dataset.md

+34-44
Original file line numberDiff line numberDiff line change
@@ -72,26 +72,6 @@ python scripts/gen_info_rope3d.py
7272
## 3. KITTI Dataset
7373
#### 3.1. Download KITTI dataset from official [website](https://www.cvlibs.net/datasets/kitti/eval_object.php?obj_benchmark=3d).
7474

75-
The directory will be as follows.
76-
```
77-
BEVHeight_Plus
78-
├── data
79-
| └── kitti
80-
| | ├── training
81-
| | | ├── calib
82-
| | | ├── label_2
83-
| | | └── images_2
84-
| | ├── testing
85-
| | | ├── calib
86-
| | | ├── label_2
87-
| | | └── images_2
88-
| | └── ImageSets
89-
| | ├── train.txt
90-
| | └── val.txt
91-
| | └── test.txt
92-
| |...
93-
├── ...
94-
```
9575
#### 3.2. Prepare infos for KITTI dataset.
9676
```
9777
python scripts/gen_info_kitti.py --data_root data/kitti
@@ -100,34 +80,44 @@ python scripts/gen_info_kitti.py --data_root data/kitti
10080
## 4. KITTI-360 Dataset
10181
#### 4.1. Download KITTI-360 dataset from official [website](https://www.cvlibs.net/datasets/kitti-360/).
10282
#### 4.2. Download the processed KITTI-360 `train_val` and dummy `testing` [labels](https://drive.google.com/file/d/1h1VmHNdoIKRecJKANt1Wj_-nDNX_HCQG/view?usp=sharing). Extract them.
103-
#### 4.3. Arrange datasets as
104-
```
105-
BEVHeight_Plus
106-
├── data
107-
| └── kitti-360
108-
| | ├── training
109-
| | | ├── calib
110-
| | | ├── label_2
111-
| | | └── images_2
112-
| | └── ImageSets
113-
| | ├── train.txt
114-
| | └── val.txt
115-
| | └── val.txt
116-
| |...
117-
├── ...
118-
```
119-
#### 4.4. Convert the dataset to KITTI format.
83+
84+
#### 4.3. Convert the dataset to KITTI format.
12085
```
121-
python data/kitti_360/calib_converter.py
122-
python data/kitti_360/bbox_converter.py
86+
python data/kitti-360/calib_converter.py
87+
python data/kitti-360/bbox_converter.py
12388
12489
# creat soft link
125-
ln -s data/kitti_360/train_val data/kitti_360/training
126-
ln -s data/kitti_360/training/label_2_converted data/kitti_360/training/label_2
90+
ln -s data/kitti-360/train_val data/kitti-360/training
91+
ln -s data/kitti-360/training/label_2_converted data/kitti-360/training/label_2
12792
```
128-
129-
130-
#### 4.5. Prepare infos for KITTI-360 dataset.
93+
The directory will be as follows.
94+
```
95+
BEVHeight_Plus
96+
├── data
97+
│ └── kitti-360
98+
│ ├── ImageSets
99+
│ ├── KITTI-360
100+
│ │ ├── calibration
101+
│ │ ├── data_2d_raw
102+
│ │ ├── data_2d_semantics
103+
│ │ ├── data_3d_boxes
104+
│ │ └── data_poses
105+
│ ├── train_val
106+
│ │ ├── calib
107+
│ │ ├── label
108+
│ │ └── images_2
109+
│ ├── training
110+
│ │ ├── calib
111+
│ │ ├── label
112+
│ │ └── images_2
113+
│ └── testing
114+
│ ├── calib
115+
│ ├── label
116+
│ └── images_2
117+
│ ...
118+
```
119+
120+
#### 4.4. Prepare infos for KITTI-360 dataset.
131121
```
132122
python scripts/gen_info_kitti.py --data_root data/kitti-360
133123
```

0 commit comments

Comments
 (0)