Looking for the Devil in the Details: Learning Trilinear Attention Sampling Network for Fine-Grained Image Recognition

By Heliang Zheng, Jianlong Fu, Zhengjun Zha, Jiebo Luo. 2019

436 citations recorded by the source corpus.

Learning subtle yet discriminative features (e.g., beak and eyes for a bird) plays a significant role in fine-grained image recognition. Existing attention-based approaches localize and amplify significant parts to learn fine-grained details, which often suffer from a limited number of parts and heavy computational cost. In this paper, we propose to learn such fine-grained features from hundreds of part proposals by Trilinear Attention Sampling Network (TASN) in an efficient teacher-student manner. Specifically, TASN consists of 1) a trilinear attention module, which generates attention maps by modeling the inter-channel relationships, 2) an attention-based sampler which highlights attended parts with high resolution, and 3) a feature distiller, which distills part features into an object-level feature by weight sharing and feature preserving strategies. Extensive experiments verify that TASN yields the best performance under the same settings with the most competitive approaches, in iNaturalist-2017, CUB-Bird, and Stanford-Cars datasets.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Neural Prototype Trees for Interpretable Fine-grained Image Recognition (2020, neutral): “448 × 448) has shown to give better results [48, 57] with e.”
This Looks Like That, Because ... Explaining Prototypes for Interpretable Image Recognition (2020, background): “This issue may also arise with other explainability methods that show or high-light image parts, such as attention mechanisms [10, 13, 44], components [37] and other part-based models e.g. [43, 45, 46].”
Efficient Classification of Very Large Images with Tiny Objects (2021, methodology): “In the modern era of deep learning, it has been used very successfully in various problems [9,12,49,53].”
Attention mechanisms in computer vision: A survey (2021, neutral): “[97, 46, 139, 70, 128, 36, 11, 156] , [137, 143, 63, 161, 159, 6, 177, 80] , [67, 180, 10, 23, 24, 163, 155, 151] , [44, 49, 106, 64, 168, 18, 11, 36] , [48, 16, 169, 153, 174, 127, 53, 92] , [89, 21, 5, 134, 65, 171, 2, 109] , [45, 82, 157, 91, 100, 129, 146, 131] , [138, 85, 175, 76, 166, 133, 167”
Multi-branch and Multi-scale Attention Learning for Fine-Grained Visual Categorization (2020, methodology): “Therefore, [8,9,10,11] take advantage of attention mechanisms to avoid this problem.”
Fine-Grained Image Analysis With Deep Learning: A Survey (2021, background): “[87] developed a trilinear attention sampling network to”
TransFG: A Transformer Architecture for Fine-grained Recognition (2021, methodology): “Early works in localization methods rely on the annotations of parts to locate discriminative regions while recent works [16, 29, 10, 54, 49] mainly adopt region proposal networks (RPN) to propose bounding boxes which contain the discriminative regions.”
The Devil is in the Channels: Mutual-Channel Loss for Fine-Grained Image Classification (2020, neutral): “4 GoogleNet + D+A TASN (CVPR19 [51]) ResNet50 87.”
Attention Convolutional Binary Neural Tree for Fine-Grained Visual Categorization (2019, methodology): “…× 92.8 MAMC [38] ResNet-101 × 93.0 MaxEnt [10] DenseNet-161 × 93.0 WS-DAN [17] Inception v3 × 93.0 DFL-CNN [ 46] ResNet-50 × 93.1 HBPASM [40] Resnet-34 × 93.8 TASN [53] VGG-19 × 93.2 TASN [53] ResNet-50 × 93.8 DCL [7] VGG-16 × 94.1 DCL [7] ResNet-50 × 94. achieving better performance, i.e. , 87 .”
Feature Normalized Knowledge Distillation for Image Classification (2020, methodology): “Recently, KD has made huge success, some works have extended this effective idea to other application domains [13, 39, 18], others [37, 17, 24] improve the standard KD by kinds of techniques such as feature based distillation and achieve better results.”