Weakly supervised fine-grained image classification via two-level attention activation model

Xiao Ke, Yanyan Huang, Wenzhong Guo

Abstract: Highlights•A connection is made between the final classification decision and convolutional feature map to accurately locate the object.•Unique subclass features are captured on the feature map by multi-scale maximum response localization methods.•The relationship between each channel and its neighboring channels is calculated to assign the important feature channels.