Giters

jhb86253817 / PIPNet

Efficient facial landmark detector

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

some question when general the labels for train

mingliutju opened this issue · comments

commented

Hi, it is very nice work! I get some question when general the labels for train. In the paper, only one point in the score map is set 1, others are set 0, and one point in the offset map is set offset,others are set 0. In the inference phrase, how to promise all the score maps predicted are right? if there is one score map predicted is wrong, which could lead a wrong index for offset map in testing, the final landmark coordinate couldn't be right! why not set all possible offset in the gird?

commented

Hi, thanks for the question. Your suggestion is reasonable, and you may explore it if you're interested. For the current design, it also works fine because even if the classification is wrong, it usually falls at the neighboring area, and in that case, the offset part can still give correct predictions.

commented

Hi, thanks for the question. Your suggestion is reasonable, and you may explore it if you're interested. For the current design, it also works fine because even if the classification is wrong, it usually falls at the neighboring area, and in that case, the offset part can still give correct predictions.

Sorry, i test the final model (myself train), the result is not so good, even i use the predicted score map with the gt offset. when the score map predicted is wrong, the final result landmark is bad,because the integer part result is wrong , the correspond offset will be 0. maybe i should set all grid offset.

commented

Hi, thanks for the question. Your suggestion is reasonable, and you may explore it if you're interested. For the current design, it also works fine because even if the classification is wrong, it usually falls at the neighboring area, and in that case, the offset part can still give correct predictions.

Sorry, i test the final model (myself train), the result is not so good, even i use the predicted score map with the gt offset. when the score map predicted is wrong, the final result landmark is bad,because the integer part result is wrong , the correspond offset will be 0. maybe i should set all grid offset.