Channel mismatched with README.md

[YunseokJANG]

Hi, @daniel-j-h

I now tried with 'r2plus1d_34_32_ig65m' model with 487 classes (after this pull request https://github.com/moabitcoin/ig65m-pytorch/pull/30 ), but I failed with an assert. I think either this part, or the description in README (see below) is incorrect: Models fine-tuned on Kinetics have 400 classes, the plain IG65 models 487 (32 clips), and 359 (8 clips) classes.

Please check the output dimension and make some appropriate changes.

Regards,

• Yunseok

[daniel-j-h]

Your initial pull request was actually wrong - I should have double checked. This one is, too.

See the CI/CD for this in

https://github.com/moabitcoin/ig65m-pytorch/blob/02010f1cbeb2040e876ef8c30889ca3ba1a1100a/.travis.yml#L45-L48

abbreviating for visibility

  - r2plus1d_34_clip8_ig65m_from_scratch.pkl --classes 487
  - r2plus1d_34_clip8_ft_kinetics_from_ig65m.pkl --classes 400
  - r2plus1d_34_clip32_ig65m_from_scratch.pkl --classes 359
  - r2plus1d_34_clip32_ft_kinetics_from_ig65m.pkl --classes 400

Can you make a pull request

1. changing the readme example back to what it was before

2. changing the wording from

   Models fine-tuned on Kinetics have 400 classes, the plain IG65 models 487 (32 clips), and 359 (8 clips) classes.

   to

   Models fine-tuned on Kinetics have 400 classes, the plain IG65 models 487 (8 clips), and 359 (32 clips) classes.

Thanks!

[YunseokJANG]

Oh, I changed but by replacing the numbers. Do you want to flip that order?

[daniel-j-h]

Perfect, thanks!