The problem faced by GradCAM and Occlusion Sensitivity methods when dealing with images with more than 3 bands.

[BEEILAB]

Description

• Imagine we have a task for image classification in the deep learning model that takes a tensor with the shape of mnc. Here, m and n are the height and width of the input tensor respectively and c is the number of channels. The Quantus toolkit for GradCAM and Occlusion Sensitivity faces an error that originates from opencv packages and mentions that the number of bands in the input image is more than 3. Wide range of tasks work with more than 3 bands. So, maybe you can modify functions to avoid such errors! Thanks

@annahedstroem

[annahedstroem]

Hi @BEEILAB thanks for sharing this! Can you paste the code that causes this error? It is helpful to know more e.g., the type fo model (torch or tf) and the dims of your data etc. Thanks!