/seg_label_generate

Processing annotations for CULane dataset.

Primary LanguageC++

This code is for processing annotations of CULane dataset

It could:
(a) generate per-pixel labels from original annotation files.
(b) generate list files for training.

Preparation

  1. Suppose that you have downloaded the CULane dataset as described in SCNN.

Note: If you have downloaded the dataset before 16th April 2018, please update the raw annotations for train&val set as described in the dataset website.

  1. Clone and build
    git clone https://github.com/XingangPan/seg_label_generate.git
    cd seg_label_generate
    
    # if you prefer makefile
    make
    
    # or, if you prefer cmake
    mkdir build && cd build
    cmake ..

Run the code

  1. To visualize annotations, simply modify the CULane path of labelGen.sh to yours and run this script.

  2. To generate per-pixel labels, remove the '-s' in labelGen.sh and run it.

  3. To generate list files with indication of the existance of lane markings (like the train_gt.txt provided in the dataset), run listGen.sh. Also remember to modify the CULane path to yours.