Opti-CAM: Optimizing saliency maps for interpretability

Hanwei Zhang, Felipe Torres, Ronan Sicre, Yannis Avrithis, Stéphane Ayache

Abstract: Highlights•Opti-CAM combines CAM and masking-based methods, with no need for extra data, network, or training.•Compared with gradient-free methods, it finds the optimal feature map weights and is on par or faster.•We introduce a new evaluation metric, average gain (AG), to replace average increase (AI).•On a few datasets, we improve the s.o.t.a according to the most relevant classification metrics.•We shed more light into how a classifier may exploit background context.