Is augmenting reference images for local validation allowed?


I’d like to know if augmenting reference images for local validation purpose is allowed in this competition.

Thank you.

Hi separate,

I’m not exactly sure what you mean by local validation, so if you could elaborate on that it might help.

But as a general rule, if those augmented reference images are being used to update your model weights, that would not be allowed.

Let me know if that helps and feel free to clarify if I’m misunderstanding something.

Hi @dd-mike,

My question is about augmenting reference images to evaluate the trained model’s performance locally.

Thank you.

@dd-mike Could I get an answer to this post?

Hi @separate. I’m not sure we have any more clarity on the question than we did before.

Are the augmented reference images being used to update your model weights?

@dd-mike No, they are not related to the training. Can I understand it as augmenting reference images is allowed if they are not used for training?

Hi @separate.

Since it’s still not entirely clear what you’re proposing, I can only suggest reviewing the competition rules to make sure you are not doing something prohibited. This is probably the most relevant rule to your question:

  • Augmenting reference images is permitted in the inference process of generating embeddings and matching scores, so long as each reference image is used independently without any interaction with other reference images. Use of augmented reference images for any other reason, including model training, is prohibited.

What you’re proposing does sound like it might be prohibited, per the above. The safest thing to do would be to use the training set, if at all possible.

Thank you for the response, @dd-mike