BDD100K: A Large-Scale Diverse Driving Video Database

Massive. Features include:

- 100,000 HD video sequences across many different conditions

- 2D Bounding Boxes annotated on 100,000 images

- 10,000 diverse images with pixel-level annotations

Comparable to Baidu's ApolloScape dataset which seeks 1M semantic images

https://arxiv.org/abs/1803.06184

Anyone know if they will be open-sourcing their tool as well?>