Exploring the use of bezier curves for lane detection —— A baseline model.

Still Work In Progress...

Download CULane dataset here.

1. install PyTorch;
2. run the following command:

pip3 install -r requirements.txt

Command-line arguments to reproduce the result:

python3 train.py --data ./CULane --log_name baseline --pretrained_weight ./weights/resnet18-5c106cde.pth --input_size 820 295 --gpu_ids 0

Note: make sure that you've downloaded the weight for ResNet18 pre-trained on ImageNet.

Run the following command:

python3 inference.py --data ./CULane --ckpt [Path to the model checkpoint file] --input_size 820 295 --gpu_ids 0 --save_name "./video/output.avi"

1. Use Focal loss;
2. Adaptive weighting according to the curvature;
3. Data resampling;
4. ......

1. Training on Video Data;
2. Use Consistent Loss;
3. ......

And Let Me Know If You Have Any Progress on These Directions!!!