A large-scale dataset for Autonomous Driving

CARL-DATASET

Overview

 
CARL-D comprises of a large-scale stereo vision-based driving videos captured from more than 100 cities of Pakistan. As a benchmark selection, 15,000 suitable images are labelled for 2D object detection and recognition. Whereas semantic segmentation benchmark contains 2500 images have pixel-level high quality fine annotations along with 5000 coarse-annotated images which could help in enabling the deep neural networks to leverage the weakly labeled data.

Benchamark Suite

 

Semantic Segmentation

Fine Annotations & Coarse Annotations

Vehicle Detection and Recognition

Bounding Box Annotations

Road Detection

Fine Annotations

DATASETS

 
Image placeholder

Vehicle Detection & Recognition

One-stage and two-stage object detectors are evaluated on our ODR-dataset to demonstrate the effectiveness of our benchmark suite.

Key Features
  • 15,000 Images/Bbox annotations
  • 50,348 Labels
  • 25 Diverse classes
  • 2x Annotation formats i.e.,Yolo & VOC-PASCAL
  • Average 7 objects per image
Image placeholder

Semantic Segmentation

Scene Segmentation dataset is categorized into seven groups: (i) nature, (ii) infrastructure, (iii) sky, (iv) moving objects, (v) static objects, (vi) signboards, and (vii) misc.

Key Features
  • 2500 fine-annotated Images
  • 5000 coarse-annotated Images
  • 42 Diverse classes
  • COCO format Annotations
  • Polygonal Annotations
Image placeholder

Drivable Road Region Detection

A benchmark for drivable road region detection and steering angle estimation on unstructured roads to ensure autonomous driving without lanes markings.

Key Features
  • 15,000 fine-annotated Images
  • Diverse road types
  • Binary classes (i) Road , (ii) Background
  • COCO format Annotations
  • Polygonal Annotations

Downloads

This dataset is for non-commercial use only.

Download Dataset

If you consider our dataset useful, please cite the following.

CARL-D: A vision benchmark suite and large scale dataset for vehicle detection and scene segmentation

Pixel Level Segmentation Based Drivable Road Region Detection and Steering Angle Estimation Method for Autonomous Driving on Unstructured Roads

Contact

Get In Touch

Contact Details

  • Email matifbutt@outlook.com, atif.ra@must.edu.pk
  • Phone N/A
  • Fax N/A
  • Address Control, Automotive, & Robotics Lab,
    Department of CS & IT, MUST
    Mirpur, Azad Kashmir, Pakistan.