Insightface: What face detection algorithm is used in demo?

Created on 4 Jan 2019  路  6Comments  路  Source: deepinsight/insightface

Hi all,

The new demo is very impressive. I'm interested in the face detection algorithm used in the demo. Is it mtcnn with some extra process (since original mtcnn's result is a bit jumpy), or is it some other algorithm?

And is the time it take to generate this demo longer than the length of the original video (can this detection and recognition algorithm work in real time)?

Most helpful comment

Thanks for your interest.
The detector is a one-shot style detector.

When the input is 720* 1280, the system can run at ~15FPS with the large model (real-time with the small model).

All 6 comments

Thanks for your interest.
The detector is a one-shot style detector.

When the input is 720* 1280, the system can run at ~15FPS with the large model (real-time with the small model).

Are these FPS CPU or GPU-based computations?

Is this new upcoming detector? Impressive results!

@jiankangdeng Thank you for your reply. Looking forward to your paper about this detector and maybe an open source implementation of it.

BTW, is it some upgraded version of the detector from your 'Face_Detection_Alignment' repo?

Is the source code for the video example available?

When the source code for the video demo available?
Thanks.

When the source code for the video demo available?
Thanks.

Was this page helpful?
0 / 5 - 0 ratings

Related issues

lmmcc picture lmmcc  路  4Comments

weihua04 picture weihua04  路  5Comments

nmzszxsl01 picture nmzszxsl01  路  4Comments

ahkarami picture ahkarami  路  4Comments

shenriver picture shenriver  路  5Comments