ELoPE: Fine-Grained Visual Classification with Efficient Localization, Pooling and Embedding

Hanselmann, Harald; Ney, Hermann

doi:10.1109/wacv45572.2020.9093601

Cited by 33 publications

(21 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thereafter, some works locate the critical region in a weakly-supervised manner with only image labels. Typical examples include RA-CNN [11], MA-CNN [46], MGE-CNN [45], ELoPE [15] as well as DP-Net [40]. Although achieving passable results, these methods mainly focus on locating discriminative regions without considering how to integrate them into a unified concept.…”

Section: Related Workmentioning

confidence: 99%

“…As mentioned above, APC module can be treated as a data augmentation method. In the training phase, the generated images (with [11] VGG-19 85.3 MA-CNN [46] VGG-19 86.5 PA-CNN [47] VGG-19 87.8 NTS-Net [41] ResNet-50 87.5 Cross-X [30] ResNet-50 87.7 DCL [3] ResNet-50 87.8 ACNet [23] ResNet-50 88.1 AP-CNN [5] ResNet-50 88.4 S3N [6] ResNet-50 88.5 SPS [21] ResNet-50 88.7 DP-Net [40] ResNet-50 89.3 PMG [8] ResNet-50 89.6 CIN [12] ResNet-101 88.1 ELoPE [15] ResNet-101 88.5 MGE-CNN [45] ResNet-101 89.4 CAL [32] ResNet-101 90.6 FDL [28] DenseNet-161 89.1 API-Net [48] DenseNet-161 90.0 Stacked-LSTM [13] GoogLeNet 90.4…”

Section: Training and Inferencementioning

confidence: 99%

See 1 more Smart Citation

ViT-FOD: A Vision Transformer based Fine-grained Object Discriminator

Zhang¹,

Chen²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Recently, several Vision Transformer (ViT) based methods have been proposed for Fine-Grained Visual Classification (FGVC). These methods significantly surpass existing CNN-based ones, demonstrating the effectiveness of ViT in FGVC tasks. However, there are some limitations when applying ViT directly to FGVC. First, ViT needs to split images into patches and calculate the attention of every pair, which may result in heavy redundant calculation and unsatisfying performance when handling fine-grained images with complex background and small objects. Second, a standard ViT only utilizes the class token in the final layer for classification, which is not enough to extract comprehensive fine-grained information.To address these issues, we propose a novel ViT based fine-grained object discriminator for FGVC tasks, ViT-FOD for short. Specifically, besides a ViT backbone, it further introduces three novel components, i.e, Attention Patch Combination (APC), Critical Regions Filter (CRF), and Complementary Tokens Integration (CTI). Thereinto, APC pieces informative patches from two images to generate a new image so that the redundant calculation can be reduced. CRF emphasizes tokens corresponding to discriminative regions to generate a new class token for subtle feature learning. To extract comprehensive information, CTI integrates complementary information captured by class tokens in different ViT layers. We conduct comprehensive experiments on widely used datasets and the results demonstrate that ViT-FOD is able to achieve state-of-the-art performance.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Training and Inferencementioning

confidence: 99%

ViT-FOD: A Vision Transformer based Fine-grained Object Discriminator

Zhang¹,

Chen²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…[20] involves LSTMs and a Mask-RCNN that needs to be pretrained on additional data. The network optimization in [20] requires multiple complex stages and is computationally expensive (also pointed out [15] 86.5% TASN [16] 87.9% HSE [17] 88.1% S3N [18] 88.5% PAIRS [1] 89.2% Subset B: Inception-v3 [19] 89.6% Stacked LSTM [20] 90 [15] 90.3% MPN-COV [21] 93.3% TASN [16] 93.8% MGE-CNN [22] 93.9% EfficientNet [23] 94.7% AutoAugment [24] 94 in [25,26]). [19] uses Earth Mover's Distance (EMD) to measure the distance between datasets, and requires much extra data for transferring the knowledge.…”

Section: Comparison With State-of-the-artmentioning

confidence: 99%

Attention-Based Multi-Task Learning For Fine-Grained Image Classification

Liu

Wang

Mase

et al. 2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

Fine-Grained Image Classification is an inherently challenging task because of its inter-class similarity and intra-class variance. Most existing studies solve this problem by localizationand-classification strategies, which, however, always causes the problem of information loss or heavy computational expenses. Instead of localization-and-classification strategy, we propose a novel end-to-end optimization procedure named Multi-Task Attention Learning (MTAL), which reinforces the neural network' correspondence to attention regions. Experimental results on CUB-Birds and Stanford Cars show that our procedure distinctly outperforms the baselines and is comparable with state-of-the-art studies despite its simplicity * .

show abstract

“…ELoPE: Fine-Grained Visual Classifcation with Effcient Localization, Pooling ELoPE 2019 93.5 and Embedding [31] Fine-Grained Visual Classifcation via Progressive Multi-Granularity Training of PMG 2020 93.4 Jigsaw Patches [32] Channel Interaction Networks for Fine-Grained Image Categorization [33] CIN 2020 93.3…”

Section: B Implementationmentioning

confidence: 99%

Improved Attentive Pairwise Interaction (API-Net) for Fine-Grained Image Classification

Yet

Rassem

Rahman

et al. 2021

2021 Emerging Technology in Computing, Communication and Electronics (ETCCE)

View full text Add to dashboard Cite

I m pr o v e d Att e nti v e P air wis e I nt er a cti o n ( A PI-N et) f or Fi n e-Gr ai n e d I m a g e Cl assi f c ati o n E m ail: m u st afi z ur @ u m p. e d u. m y O n g Z u Yet 1 , Ta h a H. R ass e m 2 , M d Ar af at ur R a h m a n 3 , a n d M. M. R a h m a n 4 1 F a c ult y of C o m p uti n g, C oll e g e of C o m p uti n g a n d A p pli e d S ci e n c es, U ni v ersiti M al a ysi a P a h a n g, P e k a n, M al a ysi a E m ail: z u . y . o n g 5 2 8 @ g m a i l . c o m 2 F a c ult y of S ci e n c e a n d Te c h n ol o g y, B o ur n e m o ut h U ni v ersit y, P o ol e B H

show abstract

ELoPE: Fine-Grained Visual Classification with Efficient Localization, Pooling and Embedding

Cited by 33 publications

References 24 publications

ViT-FOD: A Vision Transformer based Fine-grained Object Discriminator

ViT-FOD: A Vision Transformer based Fine-grained Object Discriminator

Attention-Based Multi-Task Learning For Fine-Grained Image Classification

Improved Attentive Pairwise Interaction (API-Net) for Fine-Grained Image Classification

Contact Info

Product

Resources

About