SPGC/DuckietownRoadMarkupsSegmentation

Duckietown road markups dataset

This dataset was collected under various lighting conditions, including natural daylight and different types of artificial lamp illumination. Initial annotations were generated using color ranges specifically adjusted for each distinct lighting condition during data collection.

Key observations from the collection phase:

• Annotated features located in distant parts of images appeared significantly dimmer compared to features closer to the robot.
• Yellow and red regions were particularly susceptible to glare.

Due to these issues, the top part of each image was cropped. The resulting segmentation task utilizes only the bottom 160 pixels of each image, corresponding to approximately 33 cm in front of the robot (equivalent to two Duckiebot lengths).

The final dataset comprises approximately 30,000 images.

Dataset Download and Setup

To download and set up the dataset, use the following commands:

git clone <placeholder>
python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt

python dataset.py <path_where_to_save_dataset> [experimental]

Dataset Structure

After downloading, the dataset will have the following structure at the specified path:

path_where_to_save_dataset/
├── data/          # Original images (jpg format)
└── masks/         # Pixel-wise segmentation masks (png format)

The masks directory contains segmentation masks corresponding to the bottom 160 pixels of each image, with classes encoded as follows:

• [0, 0, 0] - Background or road surface
• [255, 0, 0] - Red lane marking
• [255, 255, 255] - White lane marking
• [255, 255, 0] - Yellow lane marking

Experimental Dataset

Adding the optional flag experimental:

python dataset.py <path_where_to_save_dataset> experimental

will also unpack an additional set of approximately 10,000 images. These images contain lower-quality annotations and significantly varied lighting conditions compared to the primary dataset. This additional set can potentially improve the robustness of algorithms trained on the data, but it may introduce annotation noise due to its lower annotation quality.

Use this dataset cautiously to balance improved generalization against potential annotation noise.

Citation

If you use this dataset in your research or work, please cite our paper. Link to paper will be added as soon as it is published.