We are releasing the first 1,000 images of our internal comma10k dataset. After we clean up these labels, we'll release more. Learn more from the Medium post, or on the comma.ai discord in the #comma-pencil channel.

It's 1,000 pngs of real driving captured from the comma fleet. It's MIT license, no academic only restrictions or anything. It also includes our internal segnet's guess at category.

Run

./viewer.py

 imgs/  -- The png image files
 masks/ -- PNG segmentation masks (update these!)
 segs/  -- The outputs in probablity from our internal segnet (unreleased, too big)

 0 - #ffffff - empty
 1 -         - sky (deprecated, now undrivable)
 2 - #402020 - road (all parts, including shoulders, don't include private driveways but include public)
 3 - #ff0000 - lane markings (don't include non lane markings like turn arrows and crosswalks)
 4 - #808060 - undrivable
 5 - #00ff66 - movable (split into vehicles and people/animals?, actually don't)
 6 -         - signs and traffic lights (deprecated, now undrivable)
 7 - #cc00ff - my car (and anything inside it, including wires, mounts, etc...)

Start labelling!

Useful label tools:

• The included comma pencil tool
• An external image manipulation tool such as GIMP (Free) or Adobe Photoshop (Paid)

1. Fork this repository to your account using the "Fork" button in the top right
2. Clone your fork, and use your labelling tool of choice to label some images
3. Open a pull request to the official repository to submit your changes!

pip install Flask
./label.sh

Then open a browser window to https://localhost:5000/