Joint-Cascade-Face-Detection-and-Alignment icon indicating copy to clipboard operation
Joint-Cascade-Face-Detection-and-Alignment copied to clipboard

Published 20 hours ago • verified publisher icon kensun0

does the code lack some functions?

Open weiyt2014 opened this issue 9 years ago • 9 comments

test and train code seems nothing?

weiyt2014 avatar Sep 10 '15 08:09 weiyt2014

I do not implement train and test as functions. But you can find the similar codes in main.cpp.

kensun0 avatar Sep 10 '15 09:09 kensun0

LBFRegressor.cpp Predict function binfeatures is loss,if use binfeatures = DeriveBinaryFeat2(RandomForest_[stage1], images, image_index, current_shapes, bounding_boxs, result_face, score) ,the parameters number is wrong ,7parameters,but DeriveBinaryFeat2 is 9parameters. What's more, I hope to learn you detection and alignment use how long time! The paper show 30ms ,you receive it ?

weiyt2014 avatar Sep 11 '15 01:09 weiyt2014

  1. according to LBFRegressor.cpp, line 887. binfeatures=DeriveBinaryFeat2(RandomForest_[stage],images,image_index,current_shapes,bounding_boxs,face,score,fcount,fface);
  2. This version is used for training. i have another implementation which has been optimized to test, it coulde be faster than the paper by doing some optimizations.

PS: the speed is influenced by the mean number of weak classifiers(reject a negative sample), size of sliding window, offset of sliding, etc. I do not konw these parametesrs of origianl paper.

kensun0 avatar Sep 11 '15 04:09 kensun0

Oh,I see you have two predict functions in LBFRegressor.cpp.The mistake is founded in vector<Mat_ > LBFRegressor::Predict(const vector<Mat_ >& images, const vector<BoundingBox>& bounding_boxs, const vector<Mat_ >& ground_truth_shapes, int initial_num,vector& result_face) in the function you don't use DeriveBinaryFeat2(),binfeatures aren't initialized!

weiyt2014 avatar Sep 11 '15 05:09 weiyt2014

yes, the one you mentiond never be called. maybe i should delete it.

kensun0 avatar Sep 11 '15 06:09 kensun0

ok，is LBF6.model and Regressor6.model the same as the code 3000FPS? And the new model is inclde detection and alignment ? dou you have the trained model ?And dou you have a test code link? My Email is [email protected] .If possible ,I hope to have a chance to test the test code ! Thanks very much!

weiyt2014 avatar Sep 11 '15 06:09 weiyt2014

sorry, it is a long time since i finish it, i miss the model file.
you can train one with your dataset.

kensun0 avatar Sep 11 '15 07:09 kensun0

is LBF6.model and Regressor6.model the same as the code 3000FPS?

weiyt2014 avatar Sep 14 '15 00:09 weiyt2014

i do not remeber, you can veirify it by yourself.

kensun0 avatar Sep 14 '15 04:09 kensun0