Computer vision - ECCV 2022 . Part XXIII : 17th European Conference, Tel Aviv, Israel, October 23-27, 2022 : proceedings / / Shai Avidan [and four others]

• Pubblicazione: Cham, Switzerland : , : Springer, , [2022]; ©2022
• Descrizione fisica: 1 online resource (820 pages)
• Disciplina: 006.37
• Soggetto topico: Computer vision; Pattern recognition systems
• Persona (resp. second.): AvidanShai
• Nota di contenuto: Intro -- Foreword -- Preface -- Organization -- Contents - Part XXIII -- Accelerating Score-Based Generative Models with Preconditioned Diffusion Sampling -- 1 Introduction -- 2 Related Work -- 3 Preliminary -- 4 Method -- 4.1 Steady-State Distribution Analysis -- 4.2 Preconditioned Diffusion Sampling -- 4.3 Instantiation of Preconditioned Diffusion Sampling -- 5 Experiments -- 6 Conclusion -- References -- Learning to Generate Realistic LiDAR Point Clouds -- 1 Introduction -- 2 Related Work -- 2.1 Point Cloud Generation -- 2.2 Deep Learning for Point Clouds -- 2.3 LiDAR Simulation -- 3 Background -- 3.1 Energy-Based Models -- 3.2 Score-Based Energy Models -- 4 Method -- 4.1 LiDAR Generation -- 4.2 Posterior Sampling -- 5 Experiments -- 5.1 Experimental Setup -- 5.2 KITTI-360 Evaluation -- 5.3 NuScenes Results -- 5.4 Posterior Sampling -- 5.5 Discussions -- 6 Conclusion -- References -- RFNet-4D: Joint Object Reconstruction and Flow Estimation from 4D Point Clouds -- 1 Introduction -- 2 Related Work -- 3 Our Method: RFNet-4D -- 3.1 Overview -- 3.2 Compositional Encoder -- 3.3 Joint Decoder -- 3.4 Joint Learning -- 4 Experiments -- 4.1 Experimental Setup -- 4.2 Results -- 4.3 Ablation Studies -- 4.4 Complexity Analysis -- 5 Discussion and Conclusion -- References -- Diverse Image Inpainting with Normalizing Flow -- 1 Introduction -- 2 Related Work -- 3 Our Approach: Flow-Fill -- 3.1 Normalizing the Conditonal Distribution of Structural Priors -- 3.2 Flow Network Design -- 3.3 Guided Texture Generation -- 4 Experiments -- 4.1 Experimental Settings -- 4.2 Performance Evaluation -- 4.3 Region-Specific Semantic Transfer -- 4.4 Analysis -- 5 Conclusion -- References -- Improved Masked Image Generation with Token-Critic -- 1 Introduction -- 2 Background -- 2.1 Non-autoregressive Generative Image Transformer. ; 2.2 Challenges in Training Non-autoregressive Transformers -- 2.3 Challenges in Sampling from Non-autoregressive Transformers -- 3 Method -- 3.1 Training the Token-Critic -- 3.2 Sampling with Token-Critic -- 3.3 Relation to Discrete Diffusion Processes -- 4 Experiments -- 4.1 Experimental Setup -- 4.2 Class-Conditional Image Synthesis -- 4.3 VQ Image Refinement -- 5 Related Work -- 6 Conclusion -- References -- TREND: Truncated Generalized Normal Density Estimation of Inception Embeddings for GAN Evaluation -- 1 Introduction -- 2 Related Work -- 2.1 GANs -- 2.2 Evaluation Metrics for GANs -- 3 Analysis of Inception Features -- 4 Proposed Method -- 5 Experiments -- 5.1 Choice of Distribution -- 5.2 Comparing Metrics Using Toy Datasets -- 5.3 Density Estimation of Real-World Datasets -- 5.4 Effectiveness for Image Disturbance -- 5.5 Evaluating Generative Models -- 5.6 Robustness to the Number of Samples -- 6 Conclusion -- References -- Exploring Gradient-Based Multi-directional Controls in GANs -- 1 Introduction -- 2 Related Work -- 3 Methods -- 3.1 Discovery of Semantically Meaningful Directions -- 3.2 Disentanglement of Attributes During Manipulation -- 4 Experiments -- 4.1 Qualitative Results -- 4.2 Quantitative Evaluation -- 4.3 Ablation Studies: Attribute Disentanglement -- 5 Discussion -- References -- Spatially Invariant Unsupervised 3D Object-Centric Learning and Scene Decomposition -- 1 Introduction -- 2 Related Work -- 3 SPAIR3D -- 3.1 Local Object Proposal and Generative Model -- 3.2 Chamfer Mixture Loss -- 3.3 Model Structure -- 3.4 Soft Boundary -- 4 Experiments -- 4.1 Simulated Datasets -- 4.2 Real Dataset -- 4.3 Ablation Study of Multi-layer PointGNN -- 4.4 Empirical Evaluation of PGD -- 5 Limitations -- 6 Conclusion and Future Work -- References -- Neural Scene Decoration from a Single Photograph -- 1 Introduction -- 2 Related Work. ; 2.1 Image Synthesis -- 2.2 Scene Modeling -- 3 Proposed Method -- 3.1 Problem Formulation -- 3.2 Architecture Design -- 3.3 Training -- 4 Experiments -- 4.1 Dataset -- 4.2 Baselines -- 4.3 Quantitative Evaluation -- 4.4 Ablation Study -- 4.5 Qualitative Evaluation -- 4.6 Setting Object Sizes -- 4.7 Layout Manipulation -- 4.8 Generalization to Real-World Data -- 4.9 User Study -- 5 Conclusion -- References -- Outpainting by Queries -- 1 Introduction -- 2 Related Work -- 2.1 Image Outpainting -- 2.2 Transformer -- 3 Methodology -- 3.1 Problem Statement -- 3.2 Hybrid Transformer Autoencoder -- 3.3 Loss Functions -- 4 Experiments -- 4.1 Datasets, Implementation and Training Details -- 4.2 Experimental Results -- 4.3 Ablation Study -- 5 Conclusion -- References -- Unleashing Transformers: Parallel Token Prediction with Discrete Absorbing Diffusion for Fast High-Resolution Image Generation from Vector-Quantized Codes -- 1 Introduction -- 2 Prior Work -- 2.1 Autoregressive Models -- 2.2 Vector-Quantized Image Models -- 2.3 Discrete Energy-Based Models -- 2.4 Discrete Denoising Diffusion Models -- 2.5 Transformers -- 3 Method -- 3.1 Sampling Globally Coherent Latents -- 3.2 Addressing Gradient Variance -- 3.3 Generating High-Resolution Images -- 3.4 Improving Code Representations -- 4 Evaluation -- 4.1 Sample Quality -- 4.2 Absorbing Diffusion -- 4.3 Reconstruction Quality -- 4.4 Sample Diversity -- 4.5 Image Editing -- 4.6 Limitations -- 5 Discussion -- 6 Conclusion -- References -- ChunkyGAN: Real Image Inversion via Segments -- 1 Introduction -- 2 Related Work -- 3 Our Approach -- 4 Evaluation -- 4.1 Fidelity of Projected Images -- 4.2 Editability of Projected Images -- 4.3 Comparison with Current State-of-the-Art -- 5 Applications -- 6 Limitations -- 7 Conclusion -- References -- GAN Cocktail: Mixing GANs Without Dataset Access -- 1 Introduction. ; 2 Related Work -- 3 Problem Formulation and Baselines -- 3.1 Problem Formulation -- 3.2 Baseline A: Training from Scratch -- 3.3 Baseline B: TransferGAN -- 3.4 Baseline C: Elastic Weight Consolidation -- 4 Our Approach: GAN Cocktail -- 4.1 First Stage: Model Rooting -- 4.2 Second Stage: Model Merging -- 5 Results -- 5.1 Choosing the Root Model -- 5.2 Applications -- 6 Limitations and Future Work -- 7 Broader Impact -- 8 Conclusions -- References -- Geometry-Guided Progressive NeRF for Generalizable and Efficient Neural Human Rendering -- 1 Introduction -- 2 Related Work -- 3 Methodology -- 3.1 Geometry-Guided Multi-view Integration -- 3.2 Density Network -- 3.3 Appearance Network -- 3.4 Geometry-Guided Progressive Rendering -- 3.5 Training -- 4 Experiments -- 4.1 Datasets and Metrics -- 4.2 Implementation Details -- 4.3 Synthesis Performance Analysis -- 4.4 Efficiency Analysis -- 4.5 Ablation Studies -- 4.6 Visualization -- 5 Conclusion -- References -- Controllable Shadow Generation Using Pixel Height Maps -- 1 Introduction -- 2 Related Work -- 3 Method -- 3.1 Hard Shadow Renderer in 2D Image -- 3.2 Pixel Height Map Estimation -- 3.3 Soft Shadow Generator -- 4 Experiments and Evaluation -- 4.1 Evaluation of HENet -- 4.2 Evaluation of SSG -- 4.3 Full System Evaluation -- 5 Discussions -- 6 Conclusion -- References -- Learning Where to Look - Generative NAS is Surprisingly Efficient -- 1 Introduction -- 2 Related Work -- 3 Architecture Generative Model -- 4 Experiments -- 4.1 Experiments on Tabular Benchmarks -- 4.2 Experiments on Surrogate Benchmarks -- 4.3 Experiments on Hardware-Aware Benchmark -- 4.4 Ablation Studies -- 5 Conclusion -- References -- Subspace Diffusion Generative Models -- 1 Introduction -- 2 Background and Related Work -- 3 Subspace Diffusion -- 3.1 Score Matching -- 3.2 Sampling -- 3.3 Image Subspaces. ; 3.4 Orthogonal Fisher Divergence -- 4 Experiments -- 5 Conclusion -- References -- DuelGAN: A Duel Between Two Discriminators Stabilizes the GAN Training -- 1 Introduction -- 2 Background -- 3 DuelGAN: A Duel Between Two Discriminators -- 3.1 Formulation -- 3.2 The Max Game of Discriminators -- 3.3 The Min Game of the Generator -- 4 Properties of DuelGAN -- 4.1 DuelGAN and Common Issues in GAN Training -- 4.2 Stability and Convergence Behavior -- 5 Experiments -- 5.1 Experiment Results on Synthetic Data -- 5.2 Experiments on Real Image Datasets -- 5.3 Duel Game as a Regularizer -- 6 Conclusion -- A The DuelGAN Algorithm -- B Omitted Proofs -- B.1 Proof of Proposition 1 -- B.2 Proof of Theorem 1 -- B.3 Proof of Theorem 4 -- B.4 Proof of Theorem 3 -- B.5 Proof of Theorem 4 -- C Experiment Details and Additional Results -- C.1 Architecture Comparison Between GAN, D2GAN and DuelGAN -- C.2 Additional Experiment Results -- C.3 Additional Experiment Details -- C.4 Ablation Study of DuelGAN -- C.5 Stability of Training -- References -- MINER: Multiscale Implicit Neural Representation -- 1 Introduction -- 2 Prior Work -- 3 MINER -- 3.1 Signal Model -- 3.2 Training MINER -- 4 Experimental Results -- 5 Conclusions -- References -- An Embedded Feature Whitening Approach to Deep Neural Network Optimization -- 1 Introduction -- 2 Related Work -- 3 Embedded Feature Whitening -- 3.1 Overview of Batch Feature Whitening -- 3.2 Drawbacks of Feature Whitening -- 3.3 Removal of Recovery and Centralization Operations -- 3.4 Formulation of Embedded Feature Whitening -- 3.5 Implementation of EFW -- 4 Experiment Results -- 4.1 Experiment Setup -- 4.2 Image Classification -- 4.3 Object Detection and Segmentation -- 4.4 Person Re-identification -- 4.5 Ablation Study -- 5 Conclusion -- References -- Q-FW: A Hybrid Classical-Quantum Frank-Wolfe for Quadratic Binary Optimization. ; 1 Introduction.
• Titolo autorizzato: Computer Vision – ECCV 2022
• ISBN: 3-031-20050-0
• Formato: Materiale a stampa
• Livello bibliografico: Monografia
• Lingua di pubblicazione: Inglese
• Record Nr.: 9910624395303321
• Lo trovi qui: Univ. Federico II
• Serie: Lecture notes in computer science.