> Is weight files are required for submission?

1. I am using Google colab single notebook for running all algorithms.
So how to upload each submission with scripts?
2. Is model weight file i.e. h5 / sav file ,could to be included in the submission zip file?
3. Is results are showing immediately in codalab page ,after upload the submission.

Posted by: Hari-Krishnan @ Sept. 15, 2020, 5:44 a.m.

Hi @Hari-Krishnan,

1. Have a look at the submission format here: https://competitions.codalab.org/competitions/26027#learn_the_details-evaluation
2. No, you do not need to submit the trained weight (due to limited storage and recent updates in codalab infrastructure). However, the script should be explicit in order to generate the training weights and prediction file.
3. The submission results will be posted on the workshop website and codalab as well. However, due to some issues in codalab submission portal, the leaderboard will not reflect the results immediately.

Thanks for your participation!

Regards,
HASOC Competition Organization Team

Posted by: amitkumarjaiswal @ Sept. 15, 2020, 11:54 a.m.

If we can't submit the weight files of the trained models. Then before evaluation and generating preditict, the model needs to be trained for a longs time on GPUs. Is that okay? Will you train the model using our scripts and then generate the predition or directly use the prediction file submitted by us?

Posted by: pankajsingh @ Sept. 16, 2020, 3:40 p.m.

We do not explicitly ask participants to upload the trained weight in submission due to limited storage for codalab competition organisers. However, in a scenario where the model trained weight requires long time to generate, we allow participants to share a public link of the trained weight in the instruction file (under code.zip) to ease the scoring process.

We also updated the instruction in "Submissions & Evaluation" page for such case.

Thanks for raising this question.

Regards,
HASOC Competition Organization Team

Posted by: amitkumarjaiswal @ Sept. 16, 2020, 4:29 p.m.
Post in this thread