Target Layer for YOLOv8-X Model

[Eric-L-Manibardo]

Hello,

thanks for your amazing work, there are not many implementations for EigenCAM.

I am currently working on a project where I am using the YOLOv8-X model and I would like to apply EigenCAM for visualizing important regions in the images. However, I am unsure about which layer would be the most appropriate to target for this purpose.

Could you please provide guidance on which target layer should be set when using EigenCAM with the YOLOv8-X model? Specifically, I am looking for the layer that would provide the most informative and accurate visualizations.

Thank you for your assistance!

Additional Context:

Model: YOLOv8-X
Framework: Ultralytics YOLO

[Eric-L-Manibardo]

When using yolov8-X, the model outputs the following

However, if I do not apply the following line of code and I keep the original image size, explanations are good again (better/ more accurate regarding using yolov8-nano, which makes sense since it is smaller)

img = cv2.resize(img, (640, 640))

Still, I would like to keep the issue open for further discussion and explanation on why the resize impacts on EigenCAM.