User Added Images

Submitting images into the comma10k may help improve the driving model for your unique area. Please follow the instructions below to capture an image from your openpilot drive videos. Then you can submit it to the github repository for inclusion in model training.

Windows with WSL (Windows Subsystem for Linux)

Setup

1. Update Windows to at least the Windows 10 Fall Creators Update (released in 2017).
   https://support.microsoft.com/en-gb/help/4028685/windows-10-get-the-update

2. Install WSL
   search 'PowerShell' in the Windows search bar
   right click on the App and select "Run as administrator"
   Type following command to install WSL
   dism.exe /online /enable-feature /featurename:Microsoft-Windows-Subsystem-Linux /all /norestart
   Press Enter
   Once complete, close the Powershell window

3. Download and install Ubuntu 1804
   https://www.microsoft.com/en-us/p/ubuntu-1804-lts/9n9tngvndl3q

4. Download and install VLC media player
   https://www.videolan.org/vlc/

5. Download specific openpilot version to windows
   Go to link https://github.com/commaai/openpilot/tree/d1d2984f366e6b541b80c5364cd1f17f817ecb02
   Click the green "Code" button
   Click "Download ZIP"
   Open the .zip file with Windows File Explorer
   Click the pink "Extract" bar at the top
   Click the "Extract all" button
   In the "Select a Destination" box, type "C:\openpilot"

6. Run Install Commands
   Go to the C:\openpilot folder in File Explorer
   Hold the Shift key and right click on the white background
   Select "Open Linux Shell here"
   Type in the following commands:
   cd tools
   Press Enter
   ./ubuntu_setup.sh
   Press Enter
   Setup will take about 10-30 minutes, then continue...
   pip3 install -r requirements.txt && pip3 install pillow

Retrieve an Image

1. Login to comma server
   Type in the following commands:
   cd ..
   Press Enter
   PYTHONPATH=/mnt/c/openpilot python3 tools/lib/auth.py
   Press Enter
   A browser window should automatically open.
   Login to the google account that is paired with your openpilot device.

2. Select an image frame
   Open a new web browser window
   Go to https://my.comma.ai/
   Download the segment video file that includes the desired image frame
   Right click on the .hevc file and select "Open with.."
   Choose "VLC media player"
   Watching the video to find the desired image
   Note: VLC does not show a timestamp for .hevc files
   Open your smartphone's stopwatch app
   Start the video from the beginning and start the stopwatch simultaneously
   Stop the stopwatch at the desired image
   Multiply the number of seconds by 30 to get the image's frame number

3. Format a Fetch Image Script Command
   search 'Notepad' in the Windows search bar and open it
   Copy then paste the following template into Notepad
   PYTHONPATH=/mnt/c/openpilot python3 tools/scripts/fetch_image_from_route.py "X" N F
   Replace the F in the template with the frame number calculated from the video
   In the my.comma.ai explorer, click "copy current segment” for the desired video
   On a new line, paste the segment info into Notepad
   Note the last number in the segment info
   Replace the N in the template with the segment info's last number
   Delete the ending number and two dashes --N from the segment info
   Replace the X in the template with the shortened segment info
   For example: your segment info would start like this
   23b69975490100e4|2020-06-29--13-54-04--3
   ...and your edited template command should look like this
   PYTHONPATH=/mnt/c/openpilot python3 tools/scripts/fetch_image_from_route.py "23b69975490100e4|2020-06-29--13-54-04" 3 600

4. Run Fetch Image Script
   Go back to the Linux Shell window
   Enter the command formatted with the segment info and frame number
   Press Enter
   You will see “got api response”
   A .png image will be created in the openpilot folder like the following example:
   uxxx_23b69975490100e4_2020-06-29--13-54-04_3_600.png

5. Rename Image File
   Go to the Community Review Spreadsheet
   Go to the "User Images" tab to see the last submitted user image number
   Pick the next number that has not already been chosen for your image
   In File Explorer, rename the uxxx prefix of the image file name
   For example, the second user image's file should look like the following:
   u002_23b69975490100e4_2020-06-29--13-54-04_3_600.png

Create a Mask

Submit to Github

---

Thanks to Edwin for leading the way - https://docs.google.com/document/d/1gO65cVNYTvvhwhrbx3-UdZ5DvoqROfvu1LpdEDto9AM/edit

Fetch Image Script - https://github.com/commaai/openpilot/commit/d1d2984f366e6b541b80c5364cd1f17f817ecb02

if you added labelled images from a comma two / EON we would accept them, grab them with FrameReader from the hevc file -geohot