search

FaceDetect / jointCascade_py

Joint Cascade Face Detection and Alignment
104 stars 59 forks source link

No Face Alignment? #9

Open weiyt2014 opened 9 years ago

weiyt2014 commented 9 years ago

I run the demo_detect.py sucessfully! But I don't find 68 landmarks in detected face!And the detection time is long ,it,s 107s. in win! I want to know the code includes the alignment part? And have some good solutions to improve the detect time ? I hope to get suggestions!Thanks!

samuel1208 commented 9 years ago

Actually I don't implement the NMS algorithm. I just use cv2.grouprect to merge the candidated face rects. So I just show the face rect without landmark

weiyt2014 commented 9 years ago

I can see NMS algorithm,this algorithm calculates the area of overlap part. And compute the ratio, suppress = [suppress; pos],and it saves the max ratio. I hope to know how to connect the detection and alignment in NMS. I hope to get some suggestions.Thanks very much!

weiyt2014 commented 9 years ago

I add NMS alogrithm， but i don't see 68 landmarks! hope to get suggestions!

samuel1208 commented 9 years ago

NMS just pick one of the best rect. then show the landmark of that rect

weiyt2014 commented 9 years ago

In demo_detect.py, I use nms alogrithm to detect.The detection is better than before. But I use for i in xrange(res.shape[0]): cv2.circle(img, (res[i, 0], res[i, 1]),2, (0,0,255), -1) to show alignment,but false. And find the problem"IndexError: too many indices for array"

I hope to get some suggestions about how to realise facealignment based on your code. I see your demo_train.py including face alignment function. But I don't know how to realise it. I hope to get some suggestions! I