seg_label_generate

seg_label_generate copied to clipboard

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.

2. 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.