Pattern Recognition : 27th International Conference, ICPR 2024, Kolkata, India, December 1–5, 2024, Proceedings, Part XXV / edited by Apostolos Antonacopoulos, Subhasis Chaudhuri, Rama Chellappa, Cheng-Lin Liu, Saumik Bhattacharya, Umapada Pal.

Format:

Book

Author/Creator:

Antonacopoulos, Apostolos.

Contributor:

Chaudhuri, Subhasis.

Chellappa, Rama.

Liu, Cheng-Lin.

Bhattacharya, Saumik.

Pal, Umapada.

Series:

Lecture Notes in Computer Science, 1611-3349 ; 15325

Language:

English

Subjects (All):

Computer vision.

Machine learning.

Computer Vision.

Machine Learning.

Local Subjects:

Computer Vision.

Machine Learning.

Physical Description:

1 online resource (470 pages)

Edition:

1st ed. 2025.

Place of Publication:

Cham : Springer Nature Switzerland : Imprint: Springer, 2025.

Summary:

The multi-volume set of LNCS books with volume numbers 15301-15333 constitutes the refereed proceedings of the 27th International Conference on Pattern Recognition, ICPR 2024, held in Kolkata, India, during December 1–5, 2024. The 963 papers presented in these proceedings were carefully reviewed and selected from a total of 2106 submissions. They deal with topics such as Pattern Recognition; Artificial Intelligence; Machine Learning; Computer Vision; Robot Vision; Machine Vision; Image Processing; Speech Processing; Signal Processing; Video Processing; Biometrics; Human-Computer Interaction (HCI); Document Analysis; Document Recognition; Biomedical Imaging; Bioinformatics.

Contents:

Intro

President's Address

Preface

Organization

Contents - Part XXV

A Novel Loss for Contrastive Deep Supervision

1 Introduction

2 Related Work

2.1 Contrastive Learning

2.2 Supervised Contrastive Learning

2.3 Deep Supervision

3 Method

3.1 NCDS Framework

3.2 Analysis of .

3.3 The Novel Loss

4 Experiment

5 Conclusion

References

Multi-Task Interaction Network Based on a Cross-Attention Fusion Mechanism for Offline Signature Verification

3.1 Network Architecture

3.2 Contrastive Interaction Module

3.3 Self-Channel Interaction Module

4 Experiments

4.1 Ablation Studies

4.2 Comparison with State of the Art

4.3 Cross-Language Test

4.4 Visualization

Functional Tensor Decompositions for Physics-Informed Neural Networks

2 Theoretical Background

2.1 Universal Approximation Theorem

2.2 Physics-Informed Neural Networks

2.3 Functional Tensor Decompositions for PINNs

3 Experiments

4 Discussion and Conclusions

Squeeze and Hypercomplex Networks on Leaf Disease Detection

2 Literature Reviews

2.1 Rice Leaf Diseases Detection

2.2 Wheat Leaf Diseases Detection

2.3 Corn Leaf Diseases Detection

2.4 New Plant Leaf Diseases Data

3 Background Works

3.1 Residual 1D Convolutional Networks

3.2 Quaternion Convolution Networks

3.3 Parameterized Hypercomplex Multiplication Layer

3.4 Squeeze-and-Excitation Network

4 Proposed Squeeze-and-Hypercomplex Network

5 EXPERIMENTAL RESULTS

5.1 Dataset Description

5.2 Method

5.3 Result Analysis

5.4 Comparison with the Literature

5.5 Ablation Study

6 Conclusion

Mangoes Ripeness Grading: Vision Based Approach

1 Introduction.

2 Materials and Methods

2.1 Data Sets and Experimentation Setup

2.2 Data Augmentation

2.3 Proposed Methodology

3 Results and Discussion

4 Conclusion

FedRewind: Rewinding Continual Model Exchange for Decentralized Federated Learning

3.1 The Rewind Strategy

4 Results

4.1 Federated Learning Performance

4.2 Continual Federated Learning

On the Relationship Between Double Descent of CNNs and Shape/Texture Bias Under Learning Process

2.1 Double descent

2.2 CNN for image understanding

3 Correlation analysis framework of double descent and shape/texture bias

3.1 How to observe double descent

3.2 Phases of learning curve with double descent

3.3 Quantifying the shape/texture bias of the model

4.1 Nakkiran's setting

4.2 Ablation studies and analyses

4.3 Layer-wise analyses and visualization

5 Discussion

Cystic Adenocarcinoma Segmentation Based on Multi-frequency and Multi-scale SimAM Attention

2 Related Works

2.1 Model Architecture

2.2 Attention Mechanisms

3 Methods

3.1 Model Architecture

3.2 Fusion of Shallow Features And Deep Semantic Features Unit

3.3 Multi-Frequency in Multi-Scale SimAM Attention

4.1 Datasets

4.2 Evaluation Metrics

4.3 Experiment Settings

4.4 Comparison With SOTA Models

4.5 Ablation Study On LungSSFNet

MSDNet: A Multi-scale Dense Network for Chip Surface Defect Segmentation

2.1 Defect Classification

2.2 Defect Detection

2.3 Defect Segmentation

3.1 Architecture

3.2 Multi-scale Convolution Module

3.3 Node Module

3.4 Attention Module.

3.5 Loss Function

4.1 Dataset

4.2 Implementation Details

4.3 Experimental Results

4.4 Ablation Study

Task Oriented Image Quality Assessment for Synthesized Images

2.1 Reference-guided image synthesis (RIS)

2.2 Image Quality Assessment(IQA)

3 Methodology

3.1 Style Level Interpolation for Data Preparation

3.2 Learning-based Quality Score Estimation

3.3 Training Objective

4 EXPERIMENT

4.2 Protocol and Evalution criteria

4.3 Performance Evalution

4.4 CONCLUSION

SANGAM: Synergizing Local and Global Analysis for Simultaneous WBC Classification and Segmentation

2.1 WBC Segmentation

2.2 WBC Classification

3 Proposed System

3.1 WBC Segmentation

3.2 WBC Classification

3.3 Refined WBC Segmentation

4 Experimental Results

4.1 Experimental settings

4.2 Implementation details

4.3 Training and Testing settings

4.4 Comparative WBC Segmentation Performance

4.5 Comparative WBC Classification Performance

4.6 Ablation study

MeDiANet: A Lightweight Network for Large-scale Multi-disease Classification of Multi-modal Medical Images Using Dilated Convolution and Attention Network

3.1 Residual Block

3.2 Multi Dilated residual block

3.3 Dilated Residual Attention Block

3.4 MeDiANet

4 Experimental Setup

4.3 Evaluation Metrics

5 Results &amp

Discussions

A Data Augmentation Approach for Well Log Interpretation

2.1 Data Augmentation

2.2 Time-frequency Augmentation

3.1 Time-domain Method.

3.2 Frequency-domain Method

4.2 Experimental Setting

TSAK: Two-Stage Semantic-Aware Knowledge Distillation for Efficient Wearable Modality and Model Optimization in Manufacturing Lines

3 Apparatus and Dataset

4 Knowledge Distillation Approach

5 Result and Discussion

ArtNeRF: A Stylized Neural Field for 3D-Aware Artistic Face Synthesis

2 Related works

2.1 Style Transfer with 2D GAN

2.2 3D-aware Image Synthesis

3.1 Preliminaries

3.2 Self-supervised Style Encoder

3.3 Conditional Generative Radiance Field

3.4 Neural Rendering Module

3.5 Triple Discriminator Network

3.6 Loss Functions

4.1 Comparisons

4.2 Ablation Study

Latent Behavior Diffusion for Sequential Reaction Generation in Dyadic Setting

2.1 Deterministic reaction synthesis

2.2 Multiple Reaction Generation

3 Proposed Method

3.1 Problem definition

3.2 Facial Reaction Compression

3.3 Latent Behavior Diffusion

4.1 Evaluation setup

4.2 Evaluation metric

4.3 Results

CFTS-GAN: Continual Few-Shot Teacher Student for Generative Adversarial Networks

2 Literature Review

3.1 Continual Few-Shot GAN

3.2 Cross Domain Consistency Loss

3.3 Teacher Student Model

4.1 Qualitative Results

4.2 Quantitative Results

4.3 Ablation Study

T2R-GAN: A CGAN-based model for rural thematic road extraction

3.1 Overview of T2R-GAN.

3.2 ELAU-Net Generator and PatchGAN Discriminator

3.3 Bilateral Hinge Loss

4.1 Experimental Setting

5 Results

d-Sketch: Improving Visual Fidelity of Sketch-to-Image Translation with Pretrained Latent Diffusion Models without Retraining

3.2 Latent Code Translation Network (LCTN)

5 Conclusions

Semantically Consistent Person Image Generation

3.1 Coarse Generation Network

3.2 Data-Driven Refinement Strategy

3.3 Appearance Attribute Transfer and Rendering

6 Ablation Study

7 Limitations

8 Conclusions

GM-GAN: Geometric Generative Models Based on Morphological Equivariant PDEs and GANs

2 Equivariance and homogeneous spaces on Riemannian manifolds

3 Group morphological convolutions and PDEs

4 Morphological equivariant PDEs for generative models

4.1 Morphological PDE-based layers

4.2 PDE model design

4.3 Architecture of morphological equivariant PDEs based on GAN

5 Numerical experiments

6 Conclusion and perspectives

NR-CION: Non-rigid Consistent Image Composition Via Diffusion Model

2 Related work

2.1 Text-based image editing

2.2 Image composition

2.3 Image inversion

3 Preliminary

3.1 Latent diffusion model

3.2 Classifier free guidance

3.3 Attention mechanism

4 Method

4.1 Image inversion

4.2 Non-rigid foreground object generation and mask generation

4.3 Image composition

5 Experiments

5.1 Implementation details and benchmark

5.2 Compared with previous methods

5.3 Ablation study

6 Limitation and future work

7 Summary

References.

Neighborhood Feature Enhancement Flow Diffusion Model for Point Cloud Generation.

Notes:

Description based on publisher supplied metadata and other sources.

ISBN:

9783031783890

3031783891

OCLC:

1477220362