Coupled Patch Similarity Network FOR One-Shot Fine-Grained Image Recognition

Abstract

One-shot fine-grained image recognition (OSFG) aims to distinguish different fine-grained categories with only one training sample per category. Previous works mainly focus on learning a global feature representation through only a using single similarity metric branch, which is unsuitable for OSFG to effectively capture subtle and local differences under limited supervision. In this work, we propose a Coupled Patch Similarity Network (CPSN) for OSFG. Firstly, we propose a Feature Enhancement Module (FEM) to extract more discriminative features of the fine-grained samples. Then, we develop two coupled and symmetrical branches to capture discriminative parts of the samples and reduce the deviation of the distance metric. For each branch, we design a Patch Similarity Module (PSM) to calculate the patch similarity map for the sample pair. Especially, a Patch Weight Generator (PWG) is proposed to generate the patch weight map, which indicates the degree of importance for each position in the patch similarity map, so that the model can focus on diverse and informative parts. We analyze the effect of the different components in the proposed network, and extensive experimental results demonstrate the effectiveness and superiority of the proposed method on two fine-grained benchmark datasets.

Boosting Few-shot Fine-grained Recognition with Background Suppression and Foreground Alignment

“…Limited efforts have been devoted to distinguishing the subtle difference from the perspective of dense local features. In contrast, CPSN [49] introduces two coupled branches to compute the similarity scores between input pairs from patch level to capture subtle and local differences. FRN [50] proposes to learn a classifier by reconstructing feature maps for preserving spatial details without overfitting to pose variation.…”

A few-shot fine-grained image recognition method

“…They learn the metric model on the base of plentiful samples to spur query samples to be close to the supporting samples and generalize it to novel classes [12][13][14]. Although research [15,16] about few-shot fine-grained recognition has been carried on, the few-shot classification on fine-grained data is still a difficult problem, which is shown in Fig. 1.…”

Semantic-aware Texture-Structure Feature Collaboration for Underwater Image Enhancement

“…Recent researches reveal that semantic clues of the highlevel vision tasks [? ], [18], [29], [30] offer guidance in lowlevel vision tasks, such as super-resolution [33], [34], dehazing [26], deraining [32], and deblurring [27]. A common theme is that the semantic labels are exploited as global priors to guide networks to generate photo-realistic results.…”