search

nenadmarkus / pico

A minimalistic framework for real-time object detection (with a pre-trained face detector)
https://arxiv.org/abs/1305.4537
MIT License
641 stars 207 forks source link

can't detect rotated face #19

Open muzipeng718 opened 7 years ago

muzipeng718 commented 7 years ago

I can't detect rotated face, why?

9crk commented 7 years ago

they didnt give out the rotated model....

liberize commented 7 years ago

@9rck No need to retrain a rotated model. I managed to get rotated face detection working with the provided model facefinder. Nothing special has been done, just remember the angle argument should be in range 0.0-1.0.

9crk commented 7 years ago

@liberize can you pls tell me how to use this argument? I tried to add -a 0.0-1.0 and it's not work.

liberize commented 7 years ago

@9crk I didn't test the sample program, instead, I used the code directly in my own project. Here is some code snippets, hope it might help: https://gist.github.com/liberize/6672f53abfa871ba520c28d37365a939 PS: I modified the cluster_detections interface and added some code to get in-plane rotation angle.

liberize commented 7 years ago

I have to say, the result is not good. Some problems I have discovered:

  1. the angle calculated with maximum total quality as I did in the code above is wrong sometimes.
  2. the face rect is not accurate, sometimes too small only part of face, sometimes false positive, ...

I also tried to train a new model with part of ibug 300w dataset, still not good.

9crk commented 7 years ago

额。。都是中国人啊。。。

vvhh2002 commented 6 years ago

@liberize could you like to show us your cluster_detections code?