se-research / video-codec-performance-for-autonomous-driving

Experimental setup to systematically study the performance of video codecs for autonomous driving

video-codec-performance-for-autonomous-driving

Run in Python virtual environment (venv)

Set up virtual environment with dependencies (only has to setup once)

1. cd into video-codec-performance-for-autonomous-driving/Python
2. chmod +x install_deps.sh Makes shell script executable
3. ./install_deps.sh Runs shell script that installs all depenecies and creates venv

Adding more datasets

1. Simply put a folder containing the frames in the datasets folder (video-codec-performance-for-autonomous-driving/datasets)

NOTE The frames must be PNGs with a resolution of 2048x1536 (QXGA) (or 'KITTI' resolution - 1392x512). Every frame in each dataset must have the same resolution.

Run the script

1. source ~/py3-environments/coordinator/bin/activate Activates venv
2. python3 coordinator.py Runs script in venv:
3. deactivate Exit venv

Notes

1. In FFE.py: start.delay parameter needs to be changed according to your machine. If the delay is too low the error message 404 Client Error: Not Found ("No such container: ...") will be received for every optimization iteration.

Requirements

• Unix-like OS (tested on Arch Linux 5.0.7, Ubuntu 18.04.02 LTS and MacOS 10.14.5)
• Docker (version > 18) properly installed
• Intel QuickSync H.264 and VP9 support (Kaby Lake, Gemini Lake, Coffee Lake, Cannon Lake or later)

Acknowledgement

Christian Berger
The Revere labratory