DistilDRBA

A distilled and optimized version of DRBA featuring faster speed and improved compatibility. Providing improved speed and compatibility for real-time playback applications.

🔧Installation

git clone https://github.com/routineLife1/DistilDRBA.git
cd DistilDRBA
pip3 install -r requirements.txt

⚡Usage

Video Interpolation

  # times interpolation mode
  python infer.py -i input.mp4 -o output.mp4 -t 2
  # any fps interpolation mode
  python infer.py -i input.mp4 -o output.mp4 -fps 60

Full Usage

Usage: python infer.py -i in_video -o out_video [options]...
       
  -h                   show this help
  -w weight_path       path of the model weight (default=weights/v1.pth)
  -i input             input video path (absolute path of output video)
  -o output            output video path (absolute path of output video)
  -fps dst_fps         target frame rate (default=60)
  -t times             interpolation times (default=-1, if specified, the times mode will be used as priority)
  -s enable_scdet      enable scene change detection (default False)
  -st scdet_threshold  ssim scene detection threshold (default=0.3)
  -hw hwaccel          enable hevc_nvenc hardware acceleration encode (default False) (require nvidia graph card)
  -hf half             enable half precision to accelerate inference (default True)
  -scale scale         flow scale factor (default=1.0), generally use 1.0 with 1080P and 0.5 with 4K resolution

• weight_path = path of the model weight. Arch will be automatically selected based on the model weight path.
• input = absolute video file path. Example: E:/input.mp4
• output = absolute video file path. Example: E:/output.mp4
• dst_fps = target interpolated video frame rate. Example: 60
• times = interpolation times. Example: 2 (if specified, the times mode will be used as priority)
• enable_scdet = enable scene change detection.
• scdet_threshold = scene change detection threshold. The larger the value, the more sensitive the detection.
• hwaccel = enable hardware acceleration during encoding output video.
• half = enable half precision to accelerate inference.
• scale = flow scale factor. Decrease this value to reduce the computational difficulty of the model at higher resolutions. Generally, use 1.0 for 1080P and 0.5 for 4K resolution.

🤗 Acknowledgement

This project is supported by SVFI Development Team.