Jianwei Yang

Yes, I think that might be the reason. FRGC is 32x32, so the output feature dimesion of the network might be wrong for you 28x28 images. You can try the...

Have you solved the problem? I think you need to convert you data to 3 channels. SInce the architecture for FRGC merely takes 3 channels as input. Also, please remember...

Yes, you are right. I have implemented the testing codes using C++, which is a simple transformation from Matlab codes. :) ÔÚ 2014-11-05 20:39:14£¬"ShengyinWu" [email protected] дµÀ£º You can use C/C++...

in my laptop, it comsumes less than 1ms on c++ and a dozens of ms using matlab 发自nubia手机 twyt [email protected]编写： > How much time consume respectively on c++/c and matlab...

You can obtain the bbox_gt using any face detector as you like. the mean_shape originially is in range [0, 1; 0, 1], we need to rescale it to adapt to...

Hi, the binary tree traversing should not be that much time consuming. According to your setups, it merely needs 68 \* 6 \* 10 \* 5 times of comparison. Maybe...

@nishantkapur , it should not be that slow. I do not think the c++ implementation need to be similar to the matlab codes. Especially for feature extraction, much faster implementation...

@hao522 Hi, yes, i got some results, but the performance is worse than faster r-rcnn. Still tuning the model. Yes, we are planning extend this fpn to mask r-cnn, and...

Congratulations! If possible, hope you can share your results with others. of course you can save the final shape into pts file in your own format. It's just a simple...

Hi, @weiyt2014 , another issue should be pointed out is that you need train the face alignment model for to adapt to your own face detector. As you know, the...