Can we use TensorFlow/JAX instead of PyTorch for the final submission of the “From One to Many” challenge? Further, are we supposed to submit new x0 and x1 vectors which are conditional on the y vectors in the training set (x0_x1_y.pt)? If yes, can you please increase the max submission file size as one of the submission files I generated was 51.5MB. If no, which y vectors do we have to condition the network on.
Thank you for your questions.
To ensure proper assessment of your solution, we kindly request that you utilize the PyTorch implementation for your Final Submission.
Regarding the file size, there is no need to send us the x0/x1 vectors generated for all possible y vectors. Instead, you can select a few ( up to you 1, 10, or 30) y vectors in any manner you prefer, including random selection, for your submission.
y vectors in any manner you prefer
Hello, could we run the selection several times (in a loop) until we are happy with it (certain conditions are meet - checked by the code).
Selection could be done in any manner you prefer. However, it’s important to note that when making final submissions, you will be required to provide the full set of parameters and conditions used for the selection process.
Thank you for you creative ideas.
However, the framework desired should not include post-processing and network ensembling. Solutions obtained using these methods will not be accepted for the final evaluation stage.