Learning a Mixture of Granularity-Specific Experts for Fine-Grained Categorization

Abstract

We aim to divide the problem space of fine-grained recognition into some specific regions. To achieve this, we develop a unified framework based on a mixture of experts. Due to limited data available for the fine-grained recognition problem, it is not feasible to learn diverse experts by using a data division strategy. To tackle the problem, we promote diversity among experts by combing an expert gradually-enhanced learning strategy and a Kullback-Leibler divergence based constraint. The strategy learns new experts on the dataset with the prior knowledge from former experts and adds them to the model sequentially, while the introduced constraint forces the experts to produce diverse prediction distribution. These drive the experts to learn the task from different aspects, making them specialized in different subspace problems. Experiments show that the resulting model improves the classification performance and achieves the state-of-the-art performance on several fine-grained benchmark datasets.

Localization-Aware Adaptive Pairwise Margin Loss for Fine-Grained Image Recognition

“…Additionally, all of the results were obtained fairly, without external information such as annotations or a bounding box. As shown in Table 1, The proposed framework outperforms MGE-CNN [43], which includes many experts' input and a gating network, by 0.73% on the same ResNet-50 network. The result of the proposed method have been confirmed to be 1.13% higher than the second-highest result, ISQRT-COV [36].…”

“…Content may change prior to final publication. [33] ResNet-50 84.7 MAMC [27] ResNet-101 86.5 HBPASM [38] ResNet-34 86.8 DFL-CNN(1-scale) [35] ResNet-50 87.4 DBTNet-50 [39] ResNet-50 87.5 Cross-X [40] ResNet-50 87.7 DCL [21] ResNet-50 87.8 TASN [41] ResNet-50 87.9 iSQRT-COV [36] ResNet-50 88.1 S3N [42] ResNet-50 88.5 MGE-CNN [43] ResNet-50 88.5 MGE-CNN [43] ResNet [36] ResNet-50 92.8 MAMC [27] ResNet-101 93.0 DFL-CNN(1-scale) [35] ResNet-50 93.1 MGE-CNN [43] ResNet-101 93.6 TASN [41] ResNet-50 93.8 HBPASM [38] ResNet-34 93.8 MaxEnt [44] ResNet-50 93.85 MGE-CNN [43] ResNet-50 93.9 DBTNet-50 [39] ResNet-50 94.1 DCL [21] ResNet-50 94.5 Cross-X [40] ResNet-50…”

Exploring Misclassification Information for Fine-Grained Image Classification

“…Table 2 gives the performance comparisons of the proposed method with several baseline models [ 1 , 2 , 14 , 20 , 53 , 54 , 73 ]. We give the performances of the proposed method when combined with these baseline models.…”

“…Girshick et al [ 66 ] used the feature hierarchy while Xie et al [ 67 ] leveraged hyperclass correlation. The combination of semantic representation and multiview information were also proven effective for classification [ 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 , 78 , 79 , 80 , 81 ].…”

Text-Embedded Bilinear Model for Fine-Grained Visual Recognition

“…√ indicates YES and blank expresses NO. 82.8 Mask-CNN [44] √ 87.3 STN [19] 84.1 Bilinear-CNN [28] 84.1 Low-rank Bilinear [21] 84.2 RA-CNN [8] 85.3 HBP [49] 87.1 DFL-CNN [42] 87.4 NTS-Net [48] 87.5 DCL [6] 87.8 TASN [56] 87.9 WS-BAN [15] 88.8 Ge et al [10] 90.4 Triplet-A [7] √ Manual labour 80.7 Multi-grained [ [34] 89.5 SDR [1] 90.5 ResNet-50 [11] 92.4 RIIR [46] 94.0 NAC-CNN [38] 95.3 MGE-CNN [51] 95.9 PBC [16] 96.1 CVL [12] 96.2 TEB (ours) 96.6…”