Computational Visual Media : 12th International Conference, CVM 2024, Wellington, New Zealand, April 10–12, 2024, Proceedings, Part II / edited by Fang-Lue Zhang, Andrei Sharf.

Format:

Book

Contributor:

Zhang, Fang-Lue, editor.

Sharf, Andrei, editor.

Series:

Lecture Notes in Computer Science, 1611-3349 ; 14593

Language:

English

Subjects (All):

Computer vision.

Pattern recognition systems.

Application software.

Computer graphics.

Artificial intelligence.

Algorithms.

Computer Vision.

Automated Pattern Recognition.

Computer and Information Systems Applications.

Computer Graphics.

Artificial Intelligence.

Local Subjects:

Computer Vision.

Automated Pattern Recognition.

Computer and Information Systems Applications.

Computer Graphics.

Artificial Intelligence.

Algorithms.

Physical Description:

1 online resource (384 pages)

Edition:

1st ed. 2024.

Place of Publication:

Singapore : Springer Nature Singapore : Imprint: Springer, 2024.

Summary:

This book constitutes the refereed proceedings of CVM 2024, the 12th International Conference on Computational Visual Media, held in Wellington, New Zealand, in April 2024. The 34 full papers were carefully reviewed and selected from 212 submissions. The papers are organized in topical sections as follows: Part I: Reconstruction and Modelling, Point Cloud, Rendering and Animation, User Interations. Part II: Facial Images, Image Generation and Enhancement, Image Understanding, Stylization, Vision Meets Graphics.

Contents:

Intro

Preface

Organization

Contents - Part II

Contents - Part I

Facial Images

Zero-Shot Real Facial Attribute Separation and Transfer at Novel Views

1 Introduction

2 Related Works

2.1 Explicit Face Morphable Models

2.2 3D-Aware Implicit Models

2.3 Disentanglement Representation Learning

3 Method

3.1 Model Architecture

3.2 EM-Like Alternating Training Procedure

3.3 Model Parameters Initialization

3.4 Rendering Refinement with Blind Face Restoration

4 Experiment

4.1 Implementation Details

4.2 Zero-Shot Attribute Separation from Single Image

4.3 Comparisons

4.4 Ablation Study

5 Conclusion

5.1 Limitation

References

Explore and Enhance the Generalization of Anomaly DeepFake Detection

2 Related Work

2.1 Conventional DeepFake Detection

2.2 Anomaly DeepFake Detection

3 Approach

3.1 Overview

3.2 Review and Exploration of ADFD

3.3 Boundary Blur Mask Generator

3.4 Noise Refinement Strategy

3.5 Algorithm

4 Experiments

4.1 Experiments Setting

4.2 Exploration Experiments of ADFD Methods

4.3 Comparison Experiments

Deep Tiny Network for Recognition-Oriented Face Image Quality Assessment

2.1 Image Quality Assessment

2.2 Face Image Quality Assessment

3.1 Recognition-Oriented Non-reference Quality Measurement

3.2 Tiny Face Quality Network

3.3 Generating Training Dataset with Quality Labels

3.4 Data Sampling and Augmentation Strategy for Balancing the Distribution of Scores

4 Experimental Results

4.1 Experimental Setup

4.2 Datasets and Protocols

4.3 Visualization of Different FIQA Methods

4.4 Memory and Computation Costs

4.5 Quantitative Evaluation on IJB-B and IJB-C Datasets

4.6 Quantitative Evaluation on YTF Dataset.

4.7 Ablation Studies

Face Expression Recognition via Product-Cross Dual Attention and Neutral-Aware Anchor Loss

2.1 Landmark

2.2 Transformer in FER

2.3 Losses Used in FER

3 Our Method

3.1 Product-Cross Dual Attention Module

3.2 Neutral Expression Aware Anchor Loss

3.3 Total Loss Function

4.1 Datasets

4.2 Implementation Details

4.3 Ablation Study

4.4 Comparison with the State-of-the-Art Methods

4.5 Comparison on Number of Parameters and Running Performance

Image Generation and Enhancement

Deformable CNN with Position Encoding for Arbitrary-Scale Super-Resolution

2.1 Implicit Neural Representation

2.2 Single Image Super-Resolution (SISR)

2.3 Arbitrary-Scale Super-Resolution

3 Methods

3.1 Deformable Feature Unfolding (DFU)

3.2 Fusion with Learned Position Encoding (FPE)

3.3 Deep ResMLP

4.1 Datasets and Metrics

4.2 Implementation Detail

4.3 Evaluation

Single-Video Temporal Consistency Enhancement with Rolling Guidance

2.1 Temporal Consistency for Specific Tasks

2.2 Blind Video Temporal Consistency

2.3 Spatial Smoothing Filters and Rolling Guidance

3.2 Constructing Coarse Guidance Video

3.3 Recovering Image Details

3.4 Global Refinement

3.5 Comparison with the Deflickering Algorithm

4.1 Dataset

4.2 Quality Assessment

4.3 Comparison to State-of-the-Art Methods

5 Discussion and Conclusion

GTLayout: Learning General Trees for Structured Grid Layout Generation

3 Method.

3.1 Structural Layout Representation

3.2 Generative Model for Structured Grid Layouts

3.3 Training

4 Evaluation

4.1 Layout Generation

4.2 Layout Reconstruction

4.3 Layout Interpolation

Image Understanding

Silhouette-Based 6D Object Pose Estimation

2.1 Traditional Methods

2.2 Methods with Deep Learning

3 The Method

3.1 Problem Formulation and Notation

3.2 Dimensionality Reduction

3.3 Optimized Particle Swarm Optimization

4.1 Experiments Setup

4.2 Comparison to State of the Art

4.3 Performance on YCB-V-NT and TR-RW

4.4 Silhouette Stability Experiments

4.5 Ablation Study on YCB-V

5 Conclusion and Outlook

Robust Light Field Depth Estimation over Occluded and Specular Regions

3 The Depth Estimation

3.1 Consistency Data and Confidence

3.2 NPCR Depth Estimation

3.3 Depth Refinement

4.1 Occlusion Processing Comparisons

4.2 Specular Regions Processing

4.3 Depth Map

4.4 Computational Time

5 Conclusion and Limitation

Foreground and Background Separate Adaptive Equilibrium Gradients Loss for Long-Tail Object Detection

2.1 General Object Detection

2.2 Long-Tail Object Detection

3 Methodology

3.1 Revisiting Sigmoid Cross-Entropy Loss

3.2 Foreground and Background Separate Adaptive Equilibrium Gradients Loss

4 Experiments on LVIS

4.1 Datasets and Evaluation Metric

4.3 Ablation Studies

4.4 Generalization on Stronger Models

4.5 Performance Analysis

4.6 Comparison with State-of-the-Art Methods

4.7 Evaluation on COCO-LT

4.8 Result Visualization

Stylization.

Multi-level Patch Transformer for Style Transfer with Single Reference Image

3.1 Multi-level Patch Transformer Encoder

3.2 Dynamic Filter-Based Decoder

3.3 Loss Functions

4 Experiments and Evaluations

4.2 Qualitative Evaluation

4.4 User Study

4.5 Quantitative Evaluations

4.6 Discussion CycleTransformer vs CycleGAN

5 Conclusion and Future Work

Palette-Based Content-Aware Image Recoloring

2.1 Palette-Based Image Recoloring

2.2 Edit Propagation (Stroke-Based Image Recoloring)

2.3 Style Transfer (Example-Based Image Recoloring)

3.2 Palette Extraction

3.3 Content-Aware Recoloring

4.1 Results

4.2 Evaluation

5 Conclusion, Limitation and Future Work

FreeStyler: A Free-Form Stylization Method via Multimodal Vector Quantization

3.1 Vector Quantization Framework

3.2 Pseudo-Paired Token Predictor

4.2 Qualitative Results

4.3 Quantitative Results

4.5 Applications

5 Limitations and Future Work

6 Conclusion

Vision Meets Graphics

Denoised Dual-Level Contrastive Network for Weakly-Supervised Temporal Sentence Grounding

2.1 Weakly-Supervised Temporal Sentence Grounding

2.2 Contrastive Representation Learning

3 The Proposed Method

3.1 Problem Formulation

3.2 Visual-Text Feature Extraction

3.3 Gaussian-Based Proposal Generation

3.4 Intra-video Contrastive Learning

3.5 Inter-video Contrastive Learning

3.6 Pseudo-Label Noise Removal

3.7 Training and Inference.

4 Experiments

4.2 Evaluation Metric

4.3 Implementation Details

4.4 Comparisons with State-of-the-Art Methods

4.5 Ablation Study and Analysis

4.6 Qualitative Results

Isolation and Integration: A Strong Pre-trained Model-Based Paradigm for Class-Incremental Learning

2 Realeated Work

3.1 Problem Setting

3.2 A Simple Baseline

3.3 Dynamically Adaption and Aggregation

4.1 Experimental Setups

4.2 Comparison with State of the Art

Object Category-Based Visual Dialog for Effective Question Generation

3 Model

3.1 Object Information Extraction

3.2 Category Selection

3.3 Object Fusion Feature Update

3.4 Object-Self Difference Attention Module

3.5 Question Decoder

3.6 Object-Level Attention Update

4.2 Evaluation Metrics

4.3 Experiment Settings

4.4 Results

5 Conclusions

AST: An Attention-Guided Segment Transformer for Drone-Based Cross-View Geo-Localization

2.1 Image-Based Cross-View Geo-Localization

2.2 Vision Transformer

3 Proposed Method

3.2 Vision Transformer for Cross-View Geo-Localization

3.3 Attention-Guided Segment Tokens

3.4 Loss Function and Training Strategy

4.1 Datasets and Evaluation Metrics

4.3 Comparison with Existing Methods

4.5 Visualization

Improved YOLOv5 Algorithm for Small Object Detection in Drone Images

2.1 Object Detection

2.2 Small Object Detection

2.3 YOLOv5

3 HTH-YOLOv5

3.1 Hybrid Transformer Head.

3.2 Convolutional Attention Feature Fusion Module.

Notes:

Includes index.

ISBN:

981-9720-92-3