Hierarchical feature attention with bottleneck attention modules for multi-branch classification

While existing attention mechanisms often focus on pre-processing images, fine-grained classification tasks benefit from leveraging hierarchical relationships within categories. For example, classifying bird species involves understanding broader categories like orders and families. This inherent...

Full description

Saved in:
Bibliographic Details
Main Author: Gan, Ryan
Other Authors: Jiang Xudong
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2024
Subjects:
Online Access:https://hdl.handle.net/10356/177332
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-177332
record_format dspace
spelling sg-ntu-dr.10356-1773322024-05-31T15:44:18Z Hierarchical feature attention with bottleneck attention modules for multi-branch classification Gan, Ryan Jiang Xudong School of Electrical and Electronic Engineering EXDJiang@ntu.edu.sg Engineering While existing attention mechanisms often focus on pre-processing images, fine-grained classification tasks benefit from leveraging hierarchical relationships within categories. For example, classifying bird species involves understanding broader categories like orders and families. This inherent structure helps reduce ambiguity in predictions. This work proposes a novel approach that integrates Bottleneck Attention Mechanisms (BAM) within a ResNet50 backbone for multi-task classification. By employing separate feature branches tailored to each task and applying BAM after each branch, the model learns more discriminative features for each hierarchy. This report details the architecture and training strategy of this proposed model. Bachelor's degree 2024-05-28T00:47:51Z 2024-05-28T00:47:51Z 2024 Final Year Project (FYP) Gan, R. (2024). Hierarchical feature attention with bottleneck attention modules for multi-branch classification. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/177332 https://hdl.handle.net/10356/177332 en A3072-231 application/pdf Nanyang Technological University
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Engineering
spellingShingle Engineering
Gan, Ryan
Hierarchical feature attention with bottleneck attention modules for multi-branch classification
description While existing attention mechanisms often focus on pre-processing images, fine-grained classification tasks benefit from leveraging hierarchical relationships within categories. For example, classifying bird species involves understanding broader categories like orders and families. This inherent structure helps reduce ambiguity in predictions. This work proposes a novel approach that integrates Bottleneck Attention Mechanisms (BAM) within a ResNet50 backbone for multi-task classification. By employing separate feature branches tailored to each task and applying BAM after each branch, the model learns more discriminative features for each hierarchy. This report details the architecture and training strategy of this proposed model.
author2 Jiang Xudong
author_facet Jiang Xudong
Gan, Ryan
format Final Year Project
author Gan, Ryan
author_sort Gan, Ryan
title Hierarchical feature attention with bottleneck attention modules for multi-branch classification
title_short Hierarchical feature attention with bottleneck attention modules for multi-branch classification
title_full Hierarchical feature attention with bottleneck attention modules for multi-branch classification
title_fullStr Hierarchical feature attention with bottleneck attention modules for multi-branch classification
title_full_unstemmed Hierarchical feature attention with bottleneck attention modules for multi-branch classification
title_sort hierarchical feature attention with bottleneck attention modules for multi-branch classification
publisher Nanyang Technological University
publishDate 2024
url https://hdl.handle.net/10356/177332
_version_ 1814047253129592832