DualTopo

Checkpoint	Code and installation	Dataset / task
DualTopo SubsetA	PyTorch project	SubsetA lane segments
RATopo SubsetA	MMCV project	SubsetA 3D centerlines
RATopo SubsetB	MMCV project	SubsetB 2D centerlines

Use separate environments: DualTopo requires Python 3.11, PyTorch 2.6.0 and Transformers 4.57.3; RATopo requires Python 3.8, PyTorch 1.9.0 and MMCV-full 1.5.2.

DualTopo SubsetA

Follow the installation and data preparation instructions, then run from DualTopo/projects/dualtopo.

Three-frame setup check:

python tools/reproduce.py \
  --data-root /path/to/image_root \
  --annotations data/data_dict_subset_A_val_lanesegnet.pkl \
  --limit 3 --output-dir outputs/check_subset_a

Full evaluation on 4,806 frames:

python tools/reproduce.py \
  --data-root /path/to/image_root \
  --annotations data/data_dict_subset_A_val_lanesegnet.pkl \
  --output-dir outputs/subset_a

Weights download automatically. Both commands run init and temporal inference, then CPU scoring. Predictions and metric JSON files are saved in the output directory.

Full-validation results (%):

Branch	mAP	AP_ls	AP_ped	TOP_lsls	mAP_type
Init	36.94	35.18	38.69	28.45	50.49
Temporal	40.47	39.53	41.42	32.11	53.85

RATopo SubsetA / SubsetB

Use the shared RATopo installation and data preparation guide. Testing commands and results: SubsetA · SubsetB.

Code · License · Attribution