Download CULane and Tusimple.
Then extract them to $CULANEROOT and $TUSIMPLEROOT.
The Tusimple directory should look like:
$TUSIMPLEROOT
|──clips
|──label_data_0313.json
|──label_data_0531.json
|──label_data_0601.json
|──test_tasks_0627.json
|──test_label.json
|──readme.md
The CULane directory should look like:
$CULANEROOT
|──driver_100_30frame
|──driver_161_90frame
|──driver_182_30frame
|──driver_193_90frame
|──driver_23_30frame
|──driver_37_30frame
|──laneseg_label_w16
|──list
Our simulation dataset generator is available at anita-hu/simulanes.
Download our simulation data WATO
and unzip to $WATOROOT.
The WATO directory should look like:
$WATOROOT
|──Town01_000000.jpg
|──Town01_000000.json
|──Town01_000001.jpg
|──Town01_000001.json
|──...
To setup the TuSimple dataset with classes, download the json files from this Google Drive and place them in the TuSimple root folder. The class labels were downloaded from TuSimple-lane-classes and converted to json using the given converter script.
TuSimple has a unique structure for data and labels. There is also no segmentation labelling. TuSimple must therefore be reformatted to have a similar structure to CULane data. Additionally, segmentation labelling should be generated from the available data, to use segmentation loss.
This is accomplished by running the following script within the docker container:
python data/convert_tusimple_format.py --dataset TuSimple --root /datasets/TuSimple
This command must be run before any training or evaluation; the training and testing scripts expect the TuSimple data to already be reformatted.
Similarly the WATO data also needs to be reformatted since it uses the same label format as TuSimple
python data/convert_tusimple_format.py --dataset WATO --root /datasets/WATO_TuSimple
For image size that is not 720x1280, for example for training with CULane
python data/convert_tusimple_format.py --dataset WATO --root /datasets/WATO_CULane --res 590x1640