Feature Extraction from the conv layers

[pratikshac15]

@glenn-jocher
Hello Glenn
I have a question regarding feature extraction from the convolutional layers in YOLOv5. Specifically, I'm curious about how the model or its layers classify objects into different classes. What features does the model actually look for in the image to make these classifications? Additionally, are there reliable methods to determine exactly which features the model is using to classify objects like microplastics into specific categories?

I understand that the model may look at features like shape, texture, and color for microplastics, but I haven't found any strong references that clearly demonstrate how or if the model is truly able to "see" these particular features.

I have tried using Grad-CAM for visualization, but even then, it's still unclear which specific features the model is focusing on for the classification task. Could you provide deeper insights or suggest methods to gain a better understanding of how the model is making these classification decisions?

Thank you!