This is super-exciting, because annotated data is one of the major roadblocks to developing self-driving cars.
Prior to DeepDrive, the state-of-the-art dataset was KITTI, from the Karlsruhe Institute of Technology in Germany.
KITTI has been tremendously important for self-driving cars, and DeepDrive will build on that. For comparison, whereas KITTI has 7,500 images annotated with 3D bounding boxes, DeepDrive has 100,000. And while KITTI has 400 images for semantic segmentation, DeepDrive has 10,000.
It’s really astounding what Berkeley has released, and I’m excited to see the models people are able to build with this.