Fig 4. Analysis repeated with a smaller (64x64) and larger (256x256) patch.

A) visualization of the change in classifier output for the correct class, before the softmax activation function after occlusion by a 64x64 patch, sliding across the image in 32 pixel steps. B) The relative change in activation (compared to the original image), after occluding pixels of either the object or the background, for the different conditions (segmented, congruent, incongruent). For each image, importance values of the objects and backgrounds were normalized by dividing them by the activation for the original image, resulting in the importance ratio. Error bars represent the 95% confidence interval. C/D) Repeated for a large patch (256x256 pixels).