mean and std have the same values in the checkpoints RGB

[bguetarni]

Hi,
I downloaded the TSN (RGB) checkpoint to test it. After looking at the configuration attributes I saw that the mean and the std values for data pre-processing were actually the same. It is on purpose ?

If I do:
ckpt = torch.load("path/to/tosn_rgb.ckpt", map_location=lambda storage, loc: storage)
cfg = OmegaConf.create(ckpt["hyper_parameters"])
OmegaConf.set_struct(cfg, False)
cfg.data._root_gulp_dir = os.getcwd()
print(cfg.data.preprocessing.mean)
print(cfg.data.preprocessing.std)

Then I have:
'mean': [0.485, 0.456, 0.406]
'std': [0.485, 0.456, 0.406]

Actually after looking at the checkpoints of TRN and TMN I saw that they also have this problem.
Thanks.

[willprice]

Yes, that looks wrong to me, I'm fairly sure I didn't train them that way; good catch! I think I must have messed up the checkpoint conversion.

The standard deviation should be [0.229, 0.224, 0.225] for RGB models and the mean of that list for flow models.
Sorry about that.

I'll fix the checkpoints on dropbox by the end of next Tuesday and upload new checkpoints.

Thank you for the bug report.

[willprice]

Hi @bguetarni,
This has been rectified in d58e695 and the models have also been updated, you should find that if you redownload them (same URLs as before), then the std property will be correct.
Thanks again for raising the issue!

[willprice]

Hi @bguetarni,
It turns out I was mistaken and I did in fact train these models with the same mean and std, see #8 for more details. I've reverted the checkpoints to their original variants.