🙂 Our project code will be published soon
This repository is an official implementation of the paper 3D Lane Detection through transformers
Our methed architecture is as following
Main comparison with state-of-the-art methods on .
The dataset is divided into three parts.(1)balances scenes; (2)rarely observed; (3)visual variations
- (1)balances scenes
| Method | lane | Cneter lane | ||||||
|---|---|---|---|---|---|---|---|---|
| F-Score | AP | z-e | zz -e | F-Score | AP | z -e | zz-e | |
| 3D-LaneNet | 86.4 | 89.3 | 0.015 | 0.202 | 89.5 | 91.4 | 0.015 | 0.179 |
| Gen_LaneNet | 88.1 | 90.1 | 0.012 | 0.214 | 90.8 | 92.6 | 0.011 | 0.176 |
| ours | 92.5 | 90.4 | 0.01 | 0.239 | 95.4 | 93.3 | 0.01 | 0.219 |
- (2)rarely observed
| Method | lane | Cneter lane | ||||||
|---|---|---|---|---|---|---|---|---|
| F-Score | AP | z-e | zz -e | F-Score | AP | z -e | zz-e | |
| 3D-LaneNet | 72 | 74.6 | 0.039 | 0.521 | 77 | 80 | 0.04 | 0.557 |
| Gen_LaneNet | 78 | 79 | 0.03 | 0.539 | 79.5 | 80.6 | 0.026 | 0.547 |
| ours | 85.2 | 83.5 | 0.027 | 0.617 | 87.1 | 84.8 | 0.028 | 0.648 |
- (3)visual variations
| Method | lane | Cneter lane | ||||||
|---|---|---|---|---|---|---|---|---|
| F-Score | AP | z-e | zz -e | F-Score | AP | z -e | zz-e | |
| 3D-LaneNet | 72.5 | 74.9 | 0.032 | 0.23 | 75.5 | 77.7 | 0.03 | 0.227 |
| Gen_LaneNet | 85.3 | 87.2 | 0.015 | 0.232 | 88.2 | 90 | 0.015 | 0.187 |
| ours | 90.1 | 88.2 | 0.012 | 0.284 | 93.2 | 91.3 | 0.012 | 0.263 |
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent