For the second point：
2) Each team can submit two submissions: (1) Model with parameters between 0.2M and 2 Million maximum. (2) Model with less than 0.2M parameters --real-time upsampling. You can also submit a single one.
The competition is ranked according to PSNR. Generally speaking, contestants will submit the model with the highest score. I don't understand the significance of submitting two models？
Posted by: xiaopihai00001 @ March 7, 2025, 10:39 a.m.We will have two tables, one for efficient models, one for oofline processing.
We will acknowledge and give diplomas to the most efficient models too, since we are trying to impulse efficient ML research.
If you are a student and you do not have enough resources to train a "big" model, you can submit a small and fast model, and we will consider it.
And yes, as you might be thinking, one team can win at both tracks :)
Thank you for your detailed answer and I still have questions about the third point.
3) Models should be able to process 1MP RAWs under 5ms in regular GPUs.
1.Both two model should be able to process 1MP RAWs under 5ms?
2.what is regular GPUs?
I also have this question.
Posted by: youngyuhang @ March 18, 2025, 6:55 a.m.