Information
On this site, you can automatically remove/hide/blur faces in photos and videos. Algorithms based on neural networks are used to perform this task. You upload a media file and get result in a 1-2 minutes.
All uploaded user data is stored from one up to three days. After this time, all downloaded and created files are automatically deleted. Today we have successfully removed faces on 0 media files. You can find examples
of face bluring below or on the demo page.
Limits:
- Maximum media file size: 100 MB
- Maximum length of a video: 10 minutes
FAQ
Why hide faces at all?
The most obvious reasons are for posting sensitive videos on the internet where you don't want people to be recognized. These videos include: footage from rallies, news stories, home porn, etc.
Where can I see examples of face removal?
Examples are available on the dedicated demo page. And also below:
How "landmarks" vs "boxes" algorithms differ?
Where landmarks are used, the face is removed along the contour, that is, more accurately than when a rectangle is used to remove the face.
What types of face removal are supported?
At the moment there are 3 types: blur, black background and medium fill.
Why do you need multiple face removal algorithms?
Different algorithms give different results. Where one of the algorithms did not work well, the other can give a better result.
The following set of open repositories with different algorithms are used:
- https://github.com/timesler/facenet-pytorch
- https://github.com/ageitgey/face_recognition
- https://github.com/1adrianb/face-alignment
- https://github.com/biubug6/Pytorch_Retinaface
- https://github.com/deepcam-cn/yolov5-face
- https://github.com/elliottzheng/face-detection
What is the speed of processing for different algorithms?
FPS measurements were carried out on FullHD video on our server :
# 0 - MTCNN (boxes): Average FPS: 6.34
# 1 - DLIB (boxes): Average FPS: 4.77
# 2 - DLIB (landmarks): Average FPS: 4.77
# 3 - Face Align + SFD (landmarks, precise): Average FPS: 5.21
# 4 - Face Align + BlazeFace (landmarks, fast): Average FPS: 5.80
# 5 - RetinaFace + MobileNet (boxes, fast): Average FPS: 6.97
# 6 - RetinaFace + ResNet50 (boxes, precise): Average FPS: 2.72
# 7 - Yolov5n-Face (boxes, fast) Average FPS: 8.36
# 8 - Yolov5l-Face (boxes, precise) Average FPS: 8.12
# 0 - MTCNN (boxes): Average FPS: 6.34
# 1 - DLIB (boxes): Average FPS: 4.77
# 2 - DLIB (landmarks): Average FPS: 4.77
# 3 - Face Align + SFD (landmarks, precise): Average FPS: 5.21
# 4 - Face Align + BlazeFace (landmarks, fast): Average FPS: 5.80
# 5 - RetinaFace + MobileNet (boxes, fast): Average FPS: 6.97
# 6 - RetinaFace + ResNet50 (boxes, precise): Average FPS: 2.72
# 7 - Yolov5n-Face (boxes, fast) Average FPS: 8.36
# 8 - Yolov5l-Face (boxes, precise) Average FPS: 8.12
Donation
Neural net runs require much CPU/GPU usage, so if you want to support us, please, donate.
Contacts
If you have any questions contact "turbo [at] hide-face.com"