Modified by Jongkuk Lim
This repository was forked from lightDSFD by Jian Li
Simple implementation of video anonymization. If you are looking for a more accurate version, check Anonymizing videos by DSFD. And, if you are looking for a simpler example, noone video is implemented by only OpenCV examples.
CUDA supported enviornment
- Torch >= 0.3.1
- Torchvision >= 0.2.1
- (Tested on torch 1.3.1 and Torchvision 0.4.2)
- Python 3.6
usage: blur_video.py [-h] [--vertical VERTICAL] [--verbose VERBOSE]
[--reduce_scale REDUCE_SCALE] [--rotate ROTATE]
[--trained_model TRAINED_MODEL] [--threshold THRESHOLD]
[--cuda CUDA] [--widerface_root WIDERFACE_ROOT]
file out
positional arguments:
file Video file path
out Output video path
optional arguments:
-h, --help show this help message and exit
--vertical VERTICAL 0 : horizontal video(default), 1 : vertical video
--verbose VERBOSE Show current progress and remaining time
--reduce_scale REDUCE_SCALE
Reduce scale ratio. ex) 2 = half size of the input.
Default : 2
--rotate ROTATE Detect faces with rotation. 0 : No rotation, 1 : 90°,
2: 90°, 270°, 3 : 90°, 180°, 270°. Default : 0
--trained_model TRAINED_MODEL
Trained state_dict file path to open
--threshold THRESHOLD
Final confidence threshold
--cuda CUDA Use cuda to train model
--widerface_root WIDERFACE_ROOT
Location of VOC root directory