About experiment results

[sooonwoo]

Hi, thanks for your implementation!

Could you share the experimented results?

[a1600012888]

I can't reproduce the results in the paper due to some missing details about the hyper-param in the paper.
current results: (linear evaluation protocol on ImageNet). (also on README.md)

Train epochs	Single	Mul-5	MoCo-v2
40	55.4%	60.2%	56.9%
200	66.5%	68.3%	67.6%

[sooonwoo]

Appreciate for sharing the result!

Have you ever experimented on other datasets such as CIFAR10/100?

[a1600012888]

Thanks.
I didn't experimented on any other dataset.

[Isaver23]

Appreciate for sharing the result!

Have you ever experimented on other datasets such as CIFAR10/100?

I had some experiments on cifar10 dataset. My baseline accuracy of MoCov2 on cifar10 dataset is 90.17(linear evaluation top1 accuracy, trained with freezed backbone), and my best clsa experiment result is 91.72.

[sooonwoo]

Hi, @Isaver23

Appreciate for sharing!

[pikeyang]

Appreciate for sharing the result!
Have you ever experimented on other datasets such as CIFAR10/100?

I had some experiments on cifar10 dataset. My baseline accuracy of MoCov2 on cifar10 dataset is 90.17(linear evaluation top1 accuracy, trained with freezed backbone), and my best clsa experiment result is 91.72.

Could you share the code to train cifa10? thanks!