PyTorch implementation of
- DeepBIQ(https://arxiv.org/pdf/1602.05531.pdf)
- use DeepBIQ and ppo(https://github.com/ikostrikov/pytorch-a2c-ppo-acktr) to search usb camera image quality param(brightness, contrast, saturation, sharpness)
Contributions are very welcome. If you know how to make this code better, don't hesitate to send a pull request.
- Ubuntu 14.04
- usb camera (must support 640 * 480 resolution)
- python3.6
- PyTorch
- GPU
- LIVE In the Wild Image Quality Challenge Database (ChallengeDB_release, it must locate in: pytorch-image-quality-param-ctrl/deepbiq/ChallengeDB_release)
- if you want ppo to search good image quality param, you should add more sample to ChallengeDB_release, for example: very dark, very bright, very saturated images and so on to train DeepBIQ model
you should save your usb camera default quality param
v4l2-ctl --get-ctrl brightness
v4l2-ctl --get-ctrl contrast
v4l2-ctl --get-ctrl saturation
v4l2-ctl --get-ctrl sharpness
maybe you have to install v4l2-ctl
sudo apt-get install v4l2-ctl
cd deepbiq/
sh train.sh
cd ppo/
sh search.sh
if occurred error like:set brightness=106 failed:Input/output error ; you have to power up again usb camera