Fine-grained image recognition via trusted multi-granularity information fusion

Fine-grained image recognition (FGIR) is more challenging than general image recognition tasks due to the inherently subtle object variation. The existing FGIR methods are mainly based on single-granularity feature fusion, the extracted fused features often cannot fully reflect the characteristics of the object, and the recognition results based on the fused feature also lack interpretability. To solve this problem, we propose a novel end-to-end trusted multi-granularity information fusion (TMGIF) model for weakly-supervised fine-grained image recognition. It can automatically extract multi-granularity information representation for a fine-grained image, further evaluate the quality of information granules, and then progressively fuse multi-granularity information according to the quality to obtain a reliable and interpretable recognition result. We evaluate TMGIF on three standard benchmark datasets, and demonstrate the proposed method can provide competitive results.

Automated summary

In this paper, we propose an end-to-end trusted multi-granularity information fusion (TMGIF) model for weakly-supervised fine-grained image recognition. By automatically extracting and evaluating the quality of multi-granularity information, and progressively fusing these information, the model is able to generate reliable and interpretable recognition results.