CLGo: Learning to Predict 3D Lane Shape and Camera Pose with Geometry Constraints

PyTorch(1.9.0) training, evaluating and pretrained models for CLGo (Learning to Predict 3D Lane Shape and Camera Pose with Geometry Constraints).

• Predicting 3D lanes and camera pose from a single image.
• Learning via geometry constraints to improve performances on both tasks.

Model Zoo

We provide the CLGo model files in the .CLGoZoos/.

Set Envirionment

• Linux ubuntu 16.04
• GeForce RTX 3090
• Python 3.8.5
• CUDA 11.1

Create virtualenv environment

python3 -m venv CLGOENV

Activate it

source CLGOENV/bin/activate

Then install dependencies

pip install torch==1.9.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt

Data Preparation

Download and extract ApolloSim from yuliangguo/3D_Lane_Synthetic_Dataset

We expect the directory structure to be the following:

./CLGOENV
./CLGoZoos
./Apollo_Sim_3D_Lane_Release

Training

(1) Balanced scenes

python joint_train.py IMG_Seq_Pv-Tv_standard

(2) Rarely observed scenes

python joint_train.py IMG_Seq_Pv-Tv_rare_subset

(3) Scenes with visual variations

python joint_train.py IMG_Seq_Pv-Tv_illus_chg

Evaluation

(1) Balanced scenes

python fast_joint_test.py IMG_Seq_Pv-Tv_standard --test_mode PvTv

(2) Rarely observed scenes

python fast_joint_test.py IMG_Seq_Pv-Tv_rare_subset --test_mode PvTv

(3) Scenes with visual variations

python fast_joint_test.py IMG_Seq_Pv-Tv_illus_chg --test_mode PvTv

Acknowledgements

Gen-LaneNet