the number of samples is not 5.8M

[jetsmith]

@auroua
https://github.com/deepinsight/insightface/wiki/Dataset-Zoo
this link shows that MS1M-ArcFace has 5.8M images. But when i use mx2tfrecords.py to convert train.rec and train.idx into tfrecord, i found there are only 3804846 images. i got it by print len(imgidx) in mx2tfrecords.py. why?