Deep Attention-Based Spatially Recursive Networks for Fine-Grained Visual Recognition

doi:10.1109/TCYB.2018.2813971

Paper

Deep Attention-Based Spatially Recursive Networks for Fine-Grained Visual Recognition

Published May 1, 2019 · Lin Wu, Y. Wang, Xue Li

IEEE Transactions on Cybernetics

Q1 SJR score

213

Citations

3

Influential Citations

Full text

Semantic Scholar

Abstract

Fine-grained visual recognition is an important problem in pattern recognition applications. However, it is a challenging task due to the subtle interclass difference and large intraclass variation. Recent visual attention models are able to automatically locate critical object parts and represent them against appearance variations. However, without consideration of spatial dependencies in discriminative feature learning, these methods are underperformed in classifying fine-grained objects. In this paper, we present a deep attention-based spatially recursive model that can learn to attend to critical object parts and encode them into spatially expressive representations. Our network is technically premised on bilinear pooling, enabling local pairwise feature interactions between outputs from two different convolutional neural networks (CNNs) that correspond to distinct region detection and relevant feature extraction. Then, spatial long-short term memory (LSTMs) units are introduced to generate spatially meaningful hidden representations via the long-range dependency on all features in two dimensions. The attention model is leveraged between bilinear outcomes and spatial LSTMs for dynamic selection on varied inputs. Our model, which is composed of two-stream CNN layers, bilinear pooling, and spatial recursive encoding with attention, is end-to-end trainable to serve as the part detector and feature extractor whereby relevant features are localized, extracted, and encoded spatially for recognition purpose. We demonstrate the superiority of our method over two typical fine-grained recognition tasks: fine-grained image classification and person re-identification.

Highly Cited

Study Snapshot

Key takeawayOur deep attention-based spatially recursive model effectively identifies and classifies fine-grained objects, outperforming methods like bilinear pooling and spatial LSTMs.

PopulationOlder adults (50-71 years)

Sample size24

MethodsObservational

OutcomesBody Mass Index projections

ResultsSocial networks mitigate obesity in older groups.

Sign up to use Study Snapshot

Consensus is limited without an account. Create an account or sign in to get more searches and use the Study Snapshot.

Create an account

Paper

Deep Attention-Based Spatially Recursive Networks for Fine-Grained Visual Recognition

Sign up to use Study Snapshot

References

Exploiting Attribute Correlations: A Novel Trace Lasso-Based Weakly Supervised Dictionary Learning Method

Our weakly-supervised dictionary learning method effectively learns discriminative visual dictionary models using visual attribute correlations, improving image classification and object recognition performance.

What-and-Where to Match: Deep Spatially Multiplicative Integration Networks for Person Re-identification

Our deep multiplicative integration gating function effectively matches pedestrians across disjoint camera views, outperforming existing methods and improving person re-identification in visual recognition and surveillance.

Deep adaptive feature embedding with local sample distributions for person re-identification

Our method improves person re-identification by adaptively learning deep similarity metrics with local sample distributions, achieving local discriminations and robustness against intra-class variations.

Embedding Deep Metric for Person Re-identification: A Study Against Large Variations

Our novel moderate positive sample mining method and metric weight constraint effectively train robust CNNs for person re-identification, outperforming state-of-the-art methods on several benchmarks.

Gated Siamese Convolutional Neural Network Architecture for Human Re-identification

Our gating function improves human re-identification performance by selectively emphasizing fine common local patterns in mid-level features across pairs of images.

Citations

A Study on the Prediction Method of English Performance in Universities based on the Stacking Integrated Model

The Att-LSTM neural network long and short-term memory performance prediction model effectively predicts English performance in higher education, with enhanced dataset temporality improving the model's effectiveness.

MRGTraj: A Novel Non-Autoregressive Approach for Human Trajectory Prediction

The MRGTraj model, using a novel decoder and social context, improves human trajectory prediction performance by 13.21% and speeds-up training and testing compared to state-of-the-art models.

Exploring and Exploiting High-Order Spatial–Temporal Dynamics for Long-Term Frame Prediction

Our high-order prediction model improves long-term frame prediction accuracy by enhancing appearance and motion details through special modules.

The Image Data and Backbone in Weakly Supervised Fine-Grained Visual Categorization: A Revisit and Further Thinking

Fine-grained visual categorization (FGVC) requires considering posture, covariate shift, and structural changes, as well as differences in convolutional and self-attention networks, to develop more effective models for subclass classification.

Group Activity Representation Learning With Long-Short States Predictive Transformer

The Long-Short States Predictive Transformer (LSSPT) effectively learns group activity representations in a self-supervised manner, achieving state-of-the-art performance on Volleyball, Collective Activity, and VolleyTactic datasets.

Research on Fine-Grained Image Recognition of Birds Based on Improved YOLOv5

Our improved bird part detection algorithm based on YOLOv5 accurately recognizes bird species with 86.6% success rate, 1.2% higher than the original model.

Integrating of Digital Literacy and Achievement Motives for Learning Activities in Universities After the Covid-19 Pandemic

Integrating digital literacy and achievement motives in post-covid-19 tertiary learning activities had a low effect, highlighting the need for total changes in learning activities.