FAME generates competitive attribution maps by using gradients to drive targeted perturbations on input images rather than fixed patches, and demonstrates that CAM's locality assumption fails in deeper networks.
RISE: Random- ized input sampling for explanation of black-box models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
FAME: Feature Activation Map Explanation on Image Classification and Face Recognition
FAME generates competitive attribution maps by using gradients to drive targeted perturbations on input images rather than fixed patches, and demonstrates that CAM's locality assumption fails in deeper networks.