- Single file inference
- Folder inference
- Seeking
Tested on: python 3.9, torch 1.13, cuda 11.6, ffmpeg-python 0.2.0 (ffmpeg 4.2.2)
It assumes torch and mmsegmentation is installed. If not, you can install by following link mmsegmentation docs or commands below.
Install pre-requirement packages
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia
pip install openmin
mim install mmcv-full
pip install mmsegmentation
Install requirements.txt
pip install -r requirements.txt
config
: Path of config file
checkpoint
: Path of checkpoint file
video_path
: Directory or Filename of video file
- If video path is directory, all videos in that path will be inference.
--result_dir , -r
: Inference result save path
--interval, -i
: Inference interval frame (default=10)
--seek, -s
: Specify position of frames (e.g. -s 00:00:10 00:02:20
)
- String format following ffmepg’s format. It can be a timestamp or frame number.
Example
python3 inference_video.py -s 00:00:00 00:00:10 <config> <ckpt> <video> -r .
- Inference module outputs low resolution result with specific ffmpeg version?