Some questions regarding reproduce your amazing work

[ggosjw]

Thank you for opensourcing your great work!

I have a question regarding to reproduce your model in AV1 dataset. In your paper, you use the step learning rate schedule

But in the code you released, polyline scheduler is used:

The other question is that w in paper is set as 0.2:

But in the code you released, w is set as 0.1:

May I know which set is the correct set for me to reproduce performance of the model(simpl_av1_bezier_ckpt.tar) you released?

Thank you in advance for your kind help. Thank you again for your amazing work.