[AAAI 2026] Fine-Grained Representation for Lane Topology Reasoning

Guoqing Xu, Yiheng Li and Yang Yang

MAIS&CASIA, UCAS

Guoqing Xu and Yiheng Li contribute equally.

Introduction

This repository is an official implementation of TopoFG.

News

• [2025/11/23] Codes and weights are released.
• [2025/11/18] Camera Ready version is released.
• [2025/11/8] TopoFG is accepted by AAAI 2026 🎉🎉.

Environment Setting

• Linux
• Python 3.8.x
• NVIDIA GPU + CUDA 11.1
• PyTorch 1.9.1

conda create -n topologic python=3.8 -y
conda activate topologic

# (optional) If you have CUDA installed on your computer, skip this step.
conda install cudatoolkit=11.1.1 -c conda-forge

pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html

Install other required packages.

pip install -r requirements.txt

Data Preparation

Following OpenLane-V2 repo to download the data and run the preprocessing code.

Train & Inference

Train

The training logs will be saved to work_dirs/[work_dir_name].

cd TopoFG
mkdir work_dirs

./tools/dist_train.sh 8 [work_dir_name] [--autoscale-lr]

Evaluate

You can set --show to visualize the results.

./tools/dist_test.sh 8 [work_dir_name] [--show]

Main Results

The result is based on the v2.1.0 OpenLane-V2 devkit and metrics.

Results on OpenLane-V2 subset-A val

We provide results on Openlane-V2 subset-A val set.

Results on subset_A

Method	Epoch	OLS ↑	DETₗ ↑	DETₜ ↑	TOPₗ ↑	TOPₜ ↑
STSU	24	29.3	12.7	43.0	2.9	19.8
VectorMapNet	24	24.9	11.1	41.7	2.7	9.2
MapTR	24	31.0	17.7	43.5	5.9	15.1
TopoNet	24	39.8	28.6	48.6	10.9	23.8
TopoMLP	24	44.1	28.5	49.5	21.7	26.9
TopoLogic	24	44.1	29.9	47.2	23.9	25.4
TopoFG (Ours)	24	48.0	33.8	47.2	30.8	30.9

Weights

Method	Backbone	Epoch	Dataset	OLS	Version	Config	Download
TopoLogic	ResNet-50	24	subset-A	48.0	OpenLane-V2-v2.1.0	config	ckpt / log

Acknowledgements

We thank these great works and open-source codebases:

• TopoLogic
• TopoNet
• Openlane-V2
• MapTR, MapTRv2
• BEVFormer

Citation

If you find our work is useful, please give this repo a star and cite our work as:

@article{xu2025fine,
  title={Fine-Grained Representation for Lane Topology Reasoning},
  author={Xu, Guoqing and Li, Yiheng and Yang, Yang},
  journal={arXiv preprint arXiv:2511.12590},
  year={2025}
}