Hello, Teams and Administrator,
As I prepare my submission for this challenge, I have noted that it includes code, files, and a short paper. I have two key questions regarding this process:
(1) Must the code submitted strictly utilize the code from task1 (or task2) provided on GitHub?
I am wondering if code independently developed by participants is eligible for submission and evaluation.
(Particularly, is it mandatory for the calculation of convergence time?)
however, there are significant limitations in the usability of the GitHub code, since patch-based input is also being considered for unet optimization and training strategy which requires reconstruction when infer.
This doesn't seem to be achievable on the given github codes.
So, I wonder if it is allowed to submit code that I have written myself?
(2) I believe that the architecture of the Unet model and its parameter size can influence communication costs.
Thus, I am using an optimized model to leverage this aspect.
Will the benefits of reduced communication costs be considered in the evaluation?
Thank you.
Created by anony mous dwnsnu Hello!
(1) You must use the provided code from task 1 only with changes at the specific parts instructed. We can't accept your individually written code sadly. I understand this is limited and we hope to broaden the scope next year. We're not able to do so this year though unfortunately.
(2) For this year we don't accept changes to the model itself so we won't be able to consider this.