metadata

license: gpl-3.0
size_categories:
  - n<1K
task_categories:
  - image-to-image
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*
dataset_info:
  features:
    - name: image
      dtype: image
    - name: segment
      dtype: image
    - name: lane
      dtype: image
  splits:
    - name: train
      num_bytes: 72551321
      num_examples: 160
    - name: test
      num_bytes: 8756556
      num_examples: 20
    - name: validation
      num_bytes: 9100529
      num_examples: 20
  download_size: 90167475
  dataset_size: 90408406

About

This dataset is for detecting the drivable area and lane lines on the roads. Images are generated using stable diffusion model and images are annotated using labelme annotator.

For more info on the project we worked see this git repo

Dataset

The dataset is structured into three distinct partitions: Train, Test, and Validation. The Train split comprises 80% of the dataset, containing both the input images and their corresponding labels. Meanwhile, the Test and Validation splits each contain 10% of the data, with a similar structure, consisting of image data and label information. Within each of these splits, there are three folders:

Images: This folder contains the original images, serving as the raw input data for the task at hand.
Segments: Here, you can access the labels specifically designed for Drivable Area Segmentation, crucial for understanding road structure and drivable areas.
Lane: This folder contains labels dedicated to Lane Detection, assisting in identifying and marking lanes on the road.

Downloading the dataset

from datasets import load_dataset

dataset = load_dataset("bnsapa/road-detection")