Fig. 1
From: Refining deep convolutional features for improving fine-grained image recognition
An overview of our proposed feature extraction framework. VGG-16 [25] is used as the basic feature extractor. After rectified convolutional descriptors are extracted, they will be refined by a series of operations including multi-pooling, descriptors weighting, concatenation of two representations, and optional dimension reduction module. The panels in the figure refer to different modules in the proposed framework