This file documents a collection of baselines trained with UnLanedet. All models are trained on the AutoDL platform with a single 3090 GPU (or a 4090D GPU) with 24 GB memtory. We suggest that users train the model on this platform with the provided docker image.
We will upload the corresponding weights to facilitate the reproduction.
| Model | Backbone | Accuracy | Config | Weight | Log |
|---|---|---|---|---|---|
| SCNN | ResNet18 | 95.33 | file | - | - |
| RESA | ResNet18 | 95.62 | file | - | - |
| UFLD | ResNet18 | 94.78 | file | - | - |
| CLRNet | ResNet34 | 96.90 | file | - | - |
| LaneATT | ResNet34 | 94.65 | file | model.pth | train.log |
| ADNet | ResNet34 | 96.65 | file | model.pth | train.log |
| Model | Backbone | F1 | Config | Weight | Log |
|---|---|---|---|---|---|
| UFLD | ResNet18 | 63.14 | file | - | - |
| CLRNet | ResNet34 | 78.99 | file | model.pth | train.log |
| CondLaneNet | ResNet50 | 79.69 | file | - | - |
| CLRerNet | ResNet34 | 79.20 | file | model.pth | train.log |
| ADNet | ResNet34 | 77.88 | file | model.pth | train.log |
Note: All models are trained from scratch and reproduction results are different from the official repo.