Can I know how you suppress loss that are not used?

foreverYoungGitHub / MTCNN

Repository for "Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Neural Networks", implemented with Caffe, C++ interface.

328 stars 156 forks source link

Can I know how you suppress loss that are not used? #26

Open yxchng opened 7 years ago

yxchng commented 7 years ago

According to the MTCNN paper, "some of the losses are not used". For example, for a negative example, not bounding box and landmark points will be detected. Therefore, regression loss and landmark loss are not used. Can I know which part of your code does that? Thanks.

bpf476499866 commented 7 years ago

@foreverYoungGitHub Sorry to bother you, but I have the same question. It seems that the negative samples' roi and pts lables are all set to zero, while the original caffe-EuclideanLoss-layers, which without the ability to distinguish a sample from positive to negative, are used in train.prototxt So it meas that the negative samples will also have roi-regression loss and pts-regression loss?