Closed John1231983 closed 3 years ago
@anxiangsir is the Glint360K dataset aligned with the standard alignment method based on the retinaface detector? When do you plan to release the google drive link?
Thanks
@anxiangsir is the Glint360K dataset aligned with the standard alignment method based on the retinaface detector? When do you plan to release the google drive link?
Thanks
Glint360K has a same aligned method with MS1M-retinaface.
The dropbox link is down can you upload it again ?
Maybe you can upload to https://academictorrents.com/
@mlourencoeb
Any chance you successfully downloaded the data and can upload it again to Dropbox / Google Drive / https://academictorrents.com/ ?
Thanks
@mlourencoeb
Any chance you successfully downloaded the data and can upload it again to Dropbox / Google Drive / https://academictorrents.com/ ?
Thanks
We are fixing a bug and update the glint360k, after that, we will upload to Google Drive.
@anxiangsir It will be great, thanks!
@anxiangsir can you please let us know what the bug is? is it with the data itself?
Thanks
@facereco Yes, I was able to download it from Dropbox and the checksums seem right
@anxiangsir can you please let us know what the bug is? is it with the data itself?
Thanks
This rec supports both softmax and triplet training. There is no problem with softmax training, but triplet training has bugs.
is there an option to upload the data to a website that does not require to know fluent Chinese in order to download?
@BattashB see anxiangsir reply above: "We are fixing a bug and update the glint360k, after that, we will upload to Google Drive."
https://github.com/deepinsight/insightface/issues/1264#issuecomment-712654348
Hello, we have released glint360k's downloading magnet link, you can try it.
https://github.com/deepinsight/insightface/blob/master/recognition/partial_fc/README.md#download
@anxiangsir can you please let us know what the bug is? is it with the data itself? Thanks
This rec supports both softmax and triplet training. There is no problem with softmax training, but triplet training has bugs.
No performance drop or huge difference if using previous dataset with softmax training ? @anxiangsir
@anxiangsir can you please let us know what the bug is? is it with the data itself? Thanks
This rec supports both softmax and triplet training. There is no problem with softmax training, but triplet training has bugs.
No performance drop or huge difference if using previous dataset with softmax training ? @anxiangsir
yes
@anxiangsir can you please let us know what the bug is? is it with the data itself? Thanks
This rec supports both softmax and triplet training. There is no problem with softmax training, but triplet training has bugs.
No performance drop or huge difference if using previous dataset with softmax training ? @anxiangsir
yes
So can you explain what was wrong with dropbox file ? I already downloaded it and going to use it - is the data & labels there are all correct ?
@anxiangsir can you please let us know what the bug is? is it with the data itself? Thanks
This rec supports both softmax and triplet training. There is no problem with softmax training, but triplet training has bugs.
No performance drop or huge difference if using previous dataset with softmax training ? @anxiangsir
yes
So can you explain what was wrong with dropbox file ? I already downloaded it and going to use it - is the data & labels there are all correct ?
I saw that part in the readme but it only explains a bug in the training part (triplet loss), but not a bug in the data itself
On Wed, Oct 28, 2020, 10:25 AnXiang notifications@github.com wrote:
@anxiangsir https://github.com/anxiangsir can you please let us know what the bug is? is it with the data itself? Thanks
This rec supports both softmax and triplet training. There is no problem with softmax training, but triplet training has bugs.
No performance drop or huge difference if using previous dataset with softmax training ? @anxiangsir https://github.com/anxiangsir
yes
So can you explain what was wrong with dropbox file ? I already downloaded it and going to use it - is the data & labels there are all correct ?
see:
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/deepinsight/insightface/issues/1264#issuecomment-717777461, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABRXKAEBNOCFJR4FW6UFO2LSM7IRNANCNFSM4SNA5STQ .
Data and labels there are all correct, we only fix a bug for triplet training.
Ok thank you. I didn't know the rec file itself affects the triplet training.
On Wed, Oct 28, 2020, 10:57 AnXiang notifications@github.com wrote:
Data and labels there are all correct, we only fix a bug for triplet training.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/deepinsight/insightface/issues/1264#issuecomment-717792578, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABRXKAEA3QAGHHQBMV6BK43SM7MJJANCNFSM4SNA5STQ .
guys, did anyone uploaded dataset to gdrive? or maybe can someone please start uploading in torrent? (no peers, can't download)
please start uploading in torrent? (no peers, can't download)
Same issue here. Haven't had any peers for the past week. It would be great if the authors could re-seed Glint360K. I have a machine that could seed for the next few months.
Does anyone know a better way to share data? It's too much traffic for Google Drive .......
I can set up an FTP server for a direct file transfer if you're interested. After that, I have acces to 2 machines which could seed the torrent 24/7. Torrenting is probably the only way to avoid too much traffic to a single hosts.
Update: Mailed the details.
Hi , do you have plan to release Glint360k? If not, what is Glint360k dataset? Is Glint360k = Celeb +DeepGlint? Why not uses MS1M? Thanks