Insightface: When unpacking Glint360k rec, about 500k images are got lost.

[overpresentme]

Before unpacking rec file, the number of images (described in train.idx) was 17,507,700

But after unpacking rec file with your unpacking code, the number of images becomes 17,091,657

Is 17 million real # of images of Glint360k?

◈ Note that I downloaded Glint360k using baidu.

[anxiangsir]

This is a small bug that will not affect the usage.
The number of images is 17091657.
We will fix this bug immediately.

[yichaojin]

@anxiangsir the number of ids was about 416K (described in train.idx) not 360K, is 360K real # of ids of Glint360k?

[anxiangsir]

@anxiangsir the number of ids was about 416K (described in train.idx) not 360K, is 360K real # of ids of Glint360k?

Yes，The new rec will be finished soon, please follow us, we will update immediately

[mlourencoeb]

Deae @anxiangsir,

I am already using glint360k. These 50K difference are replicated ids?

[huanleo]

@anxiangsir the number of ids was about 416K (described in train.idx) not 360K, is 360K real # of ids of Glint360k?

Yes，The new rec will be finished soon, please follow us, we will update immediately

when i unpack the train.rec to get images, some images are corrupted, wil this be solved in the updating train.rec?

[anxiangsir]

Deae @anxiangsir,

I am already using glint360k. These 50K difference are replicated ids?

see https://github.com/deepinsight/insightface/issues/1264#issuecomment-712765364

[anxiangsir]

Hello, we have released glint360k's downloading magnet link, triplet bug has been fixed, you can try it.

https://github.com/deepinsight/insightface/blob/master/recognition/partial_fc/README.md#download