Since it's for an efficient image restoration algorithm, is there a limit to the number of parameters?
Posted by: hhyz @ May 26, 2022, 11:20 a.m.Hi, we do not set a hard limit for this challenge. But it would be better not to design too large models, say, with more than 15M trainable parameters.
Best,
Organizers
Hi~ I want to confirm is the restriction of 15M params a clear threshold? (for example, the models used for inference with > 15M params is seen as invalid submission), or it is not a clear restriction but with punishment (e.g. (actual_weight -15) as score weight), or it is just for suggestion and will not affect final submission validaty and ranking. Thanks~
Posted by: jzsherlock @ June 18, 2022, 11:37 a.m.Hi @jzsherlock,
It is a suggestion, not a hard restriction. The model size would not contribute to the final score.
Best,
Organizers