Datasets for nighttime and low-light conditions.

Object Detection

• ExDark (Exclusively Dark Image Dataset)
  -- Labelled
  -- Multiple objects are annotated.
• BDD100K: Night
  -- Labelled (Small Data)
  -- Cars and other vehicles are annotated.

Face Detection

• DarkFace Dataset
  -- Labelled
  -- 5400/600 images for train/val
  -- Face detection in extremely low-light

People/Pedestrian Detection

• NightOwls Dataset
  -- Labelled
  -- Pedestrian detection
• KAIST Multi-Spectral Dataset
  -- Labelled
  -- Pedestrian detection

Pose Estimation

• ExLPose Dataset
  -- Labelled
  -- Human 2D Pose estimation in extremely low-light

Stereo Vision

• The Oxford RobotCar
  -- Unlabelled
  -- Provides raw LIDAR based sparse depth maps.
• DENSE: Gated2Depth
  -- Unlabelled
  -- Provides raw LIDAR based sparse depth maps
• DENSE: PixelAccurateDepthBenchmark
  -- Labelled (Small Data)
  -- 4 scenes with 10 images each. Each image comes with a pixel-accurate dense depth map.

Semantic Segmentation

• BDD100K: Night
  -- Labelled (Small Data)
  -- Semantic classes compatible with the CityScapes dataset
• Dark Zurich
  -- Unlabelled
  -- Provides a small set of 50 labelled images for validation/test purposes.
  -- Semantic classes compatible with the CityScapes dataset
• NightCity
  -- Yet to be published