Open Dataset for Automotive ML Practitioners

Synthetic human data for developing ML models in automotive in-cabin CV systems.

Accurate | Performant | Ethical | Results

The Diverse Human Drivers Dataset includes synthetic humans with a range of hair styles, skin tones, epicanthic folds, clothing styles and other variables, with a fixed, in-cabin camera focused on a driver operating a vehicle. Specs:

  • 1,000 images, 500 RGB and NIR pairs
  • Tunable and varied NIR emitter intensities
  • 500 diverse identities spanning gender, skin tone, and age. 1024 x 1024 RGB images
  • Rich set of pixel-perfect labels, including segmentation maps, depth, surface normals, and 3D landmarks
  • Variation in eye gaze spanning 30° in all directions
  • Diverse clothing, hairstyles, and facial hair
  • Accessories: hats, glasses, and face masks
  • Variation in background over a wide range of lighting and confounding elements
  • Several fixed in-cabin camera angles around the driver

Access the Diverse Human Drivers Open Dataset

Please provide the following, and we will email you instructions for downloading and using the Diverse Human Drivers dataset from Synthesis AI.

ML Model Training and Development for ADAS, Driver Monitoring Systems (DMS) and Occupant Monitoring Systems (OMS)

ADAS, DMS and OMS models are trained using large amounts of data collected from various sources, including cameras and sensors mounted inside the car. Auto OEMs and software companies can augment real-world and open source training data with synthetic datasets. Models built with data augmentation offer greater accuracy, better performance, and best-in-class safety and efficiency outcomes compared with models developed using human-annotated training datasets exclusively.

Mid_breakout_ID321_3up_0000