RoadSideTrack-PyTorch

vision-based lane & side tracking algorithm (train code included) written in Python with PyTorch using fscn model.

Installation

Clone this repository:

git clone https://github.com/SheldonFung98/RoadSideTrack-PyTorch

Usage

Train

Link the checkpoint & dataset folder to the root of this repository.

ln -s /path/to/your/checkpoint_folder checkpoints
ln -s /path/to/your/datasets_folder datasets

The structure of datasets folder should be as follows:

datasets
    - cityscapes
        - gtFine
            - train
                1_gtFine_labelIds.png
                2_gtFine_labelIds.png
                ...
            - test
                ...
            - val
                ...
        - leftImg8bit
            - train
                1.png
                2.png
                ...
            - test
                ...
            - val
                ...

• The data images in leftImg8bit should be RGB image.
• The label image in gtFine should be three channel mask image.
• Configurations can be edit in configs/cityscapes_fast_scnn.yaml.

Now you can train your model with the following command:

python3 tools/train.py

Run

You can see the segmentation result via running the provided python script.

Inference with TensorRT

Please check this repo