Request for clarification on PolypMix implementation with positive and negative image mixing

[ClaudiaDelprete]

Dear author,
Thank you for sharing your work and the implementation of the PolypMix method. I’ve been exploring the code and have a question regarding the specific implementation of the mixing between positive images (containing polyps) and negative images (without polyps) as described in your paper. In the paper, Figure 8 shows examples of mixed images generated by PolypMix, where images from a normal input (without polyps) are combined with polyp images. However, in the provided implementation, it seems that PolypMix only mixes unlabeled images from the positive dataset (i.e., images with polyps).

I was wondering if the provided code supports the mixing between positive and negative images as shown in the paper. If not, could you please provide guidance on how to implement this behavior while keeping the training process coherent? Specifically:

• How should the dataset be structured to include both positive and negative images for the PolypMix method?
• Is there a mechanism in the current implementation to handle mixing between positive and negative samples, or does this require additional modifications?

Thank you.

[YChienHung]

Thanks for your attention to our work !
We only upload the code about training and evaluation, and we will upload the code you need.

Just wait a moment !

[YChienHung]

@ClaudiaDelprete Hello, I have upload the code you need.

[ClaudiaDelprete]

Thank you so much for your prompt response and for taking the time to provide the necessary code!