Bilinear Residual Attention Networks for Fine-Grained Image Classification

Yang Wang; Libo Liu

doi:10.3788/LOP57.121011

Journals >Laser & Optoelectronics Progress >Volume 57 >Issue 12 >Page 121011 > Article

Laser & Optoelectronics Progress
Vol. 57, Issue 12, 121011 (2020)

Bilinear Residual Attention Networks for Fine-Grained Image Classification

Yang Wang and Libo Liu^*

Author Affiliations

School of Information Engineering, Ningxia University, Yinchuan, Ningxia 750021, China

show less

DOI: 10.3788/LOP57.121011 Cite this Article Set citation alerts

Yang Wang, Libo Liu. Bilinear Residual Attention Networks for Fine-Grained Image Classification[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121011 Copy Citation Text

show less

Fig. 1. Architecture of B-CNN model

Download full size

Fig. 2. Schematic of bilinear feature combination

Download full size

Fig. 3. Structure of residual unit

Download full size

Fig. 4. Channel attention module

Download full size

Fig. 5. Spatial attention module

Download full size

Fig. 6. Residual attention structure of BRAN model

Download full size

Fig. 7. Example of training data augmentation

Download full size

Fig. 8. Visualization of different feature maps. (a) Original images; (b) B-CNN; (c) channel attention maps; (d) spatial attention maps

Download full size

Dataset	Class	Train	Test	Total
CUB-200-2011	200	5994	5794	11788
Stanford Dogs	120	12000	8580	20580
Stanford Cars	196	8144	8041	16185

Table 1. Detailed statistics of three fine-grained image datasets

Approach	Backbone	Accuracy /%
B-CNN(baseline)	VGG-M+VGG-D	84.1
B-CNN(resnet×2)	ResNet-34×2	85.0
BRAN(cha. attention)	ResNet-34×2 + channel attention	86.2
BRAN(spa. attention)	ResNet-34×2 + spatial attention	85.5
BRAN(cha.& spa. attention)	ResNet-34×2 + cha. & spa. attention	87.2

Table 2. Ablation experiment and analysis of proposed method on CUB-200-2011 dataset

Approach	Backbone	Accuracy /%
Approach	Backbone	Birds^[9]	Dogs^[10]	Cars^[11]
Two-level attention^[22]	VGG19	77.9	-	-
NAC^[23]	VGG19	81.01	68.61	-
B-CNN^[6]	VGG-M+VGG-D	84.1	-	91.3
ST-CNN^[24]	Inception-v2×3	84.1	-	-
DVAN^[25]	VGG-16×3	79.0	81.5	87.1
RA-CNN^[26]	VGG-19×3	85.3	87.3	92.5
MA-CNN^[19]	VGG-19×3	86.5	-	92.8
MAMC^[27]	ResNet-101	86.5	85.2	93.0
BRAN	ResNet-34×2	87.2	89.2	92.5

Table 3. Comparison with weakly-supervised methods in terms of classification accuracy

Yang Wang, Libo Liu. Bilinear Residual Attention Networks for Fine-Grained Image Classification[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121011

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information