11,939 Hits in 6.2 sec

Implicit Regularization via Neural Feature Alignment [article]

Aristide Baratin, Thomas George, César Laurent, R Devon Hjelm, Guillaume Lajoie, Pascal Vincent, Simon Lacoste-Julien
2021 arXiv   pre-print
We highlight a regularization effect induced by a dynamical alignment of the neural tangent features introduced by Jacot et al, along a small number of task-relevant directions.  ...  We approach the problem of implicit regularization in deep learning from a geometrical viewpoint.  ...  APPENDICES: Implicit Regularization via Neural Feature Alignment A Tangent Features and Geometry We describe in more formal detail some of the notions introduced in Section 2 of the paper.  ... 
arXiv:2008.00938v3 fatcat:xtcsbf4kcnbn3itixjq3ddrwhy

deepManReg: a deep manifold-regularized learning model for improving phenotype prediction from multi-modal data [article]

Nam D Nguyen, Jiawei Huang, Daifeng Wang
2021 bioRxiv   pre-print
First, deepManReg employs deep neural networks to learn cross-modal manifolds and then align multi-modal features onto a common latent space.  ...  Second, deepManReg uses cross-modal manifolds as a feature graph to regularize the classifiers for improving phenotype predictions and also prioritizing the multi-modal features and cross-modal interactions  ...  predictions via regularization by cross-modal manifolds.  ... 
doi:10.1101/2021.01.28.428715 fatcat:dp54ad5fszdjngjfmxrp3g65da

Neural Template: Topology-aware Reconstruction and Disentangled Generation of 3D Meshes [article]

Ka-Hei Hui, Ruihui Li, Jingyu Hu, Chi-Wing Fu
2022 arXiv   pre-print
This paper introduces a novel framework called DTNet for 3D mesh reconstruction and generation via Disentangled Topology.  ...  Beyond previous works, we learn a topology-aware neural template specific to each input then deform the template to reconstruct a detailed mesh while preserving the learned topology.  ...  the inverselytransformed implicit shape TI (from g −1 ) aligns with the composed implicit neural template T I (from f ).  ... 
arXiv:2206.04942v1 fatcat:ufrpzajkvza4ncci6csrfzqhca

Topology-Preserving Shape Reconstruction and Registration via Neural Diffeomorphic Flow [article]

Shanlin Sun, Kun Han, Deying Kong, Hao Tang, Xiangyi Yan, Xiaohui Xie
2022 arXiv   pre-print
Deep Implicit Functions (DIFs) represent 3D geometry with continuous signed distance functions learned through deep neural nets.  ...  The diffeomorphic deformation is realized by an auto-decoder consisting of Neural Ordinary Differential Equation (NODE) blocks that progressively map shapes to implicit templates.  ...  Neural Mesh Flow [25] focuses on generating manifold mesh from images or point clouds via conditional continuous diffeomorphic flow.  ... 
arXiv:2203.08652v2 fatcat:iwdse5kkpfh5hayoy2tgxtuzaq

Geo-PIFu: Geometry and Pixel Aligned Implicit Functions for Single-view Human Reconstruction [article]

Tong He, John Collomosse, Hailin Jin, Stefano Soatto
2020 arXiv   pre-print
Our method is based on a deep implicit function-based representation to learn latent voxel features using a structure-aware 3D U-Net, to constrain the model in two ways: first, to resolve feature ambiguities  ...  in query point encoding, second, to serve as a coarse human shape proxy to regularize the high-resolution mesh and encourage global shape regularity.  ...  These implicit methods train deep neural networks to estimate dense, continuous occupancy fields from which meshes may be reconstructed e.g. via Marching Cubes [20] .  ... 
arXiv:2006.08072v2 fatcat:7qbnswptdfapzh2e5bxygvrsgy

Coupling Explicit and Implicit Surface Representations for Generative 3D Modeling [article]

Omid Poursaeed and Matthew Fisher and Noam Aigerman and Vladimir G. Kim
2020 arXiv   pre-print
We propose a novel neural architecture for representing 3D surfaces, which harnesses two complementary shape representations: (i) an explicit representation via an atlas, i.e., embeddings of 2D domains  ...  We make these two representations synergistic by introducing novel consistency losses that ensure that the surface created from the atlas aligns with the level-set of the implicit function.  ...  Second, we observe that the gradient of the implicit representation should align with the surface normal of the explicit representation.  ... 
arXiv:2007.10294v2 fatcat:pl7yk2ucpnfonifedy5v7nediq

Iso-Points: Optimizing Neural Implicit Surfaces with Hybrid Representations [article]

Wang Yifan, Shihao Wu, Cengiz Oztireli, Olga Sorkine-Hornung
2021 arXiv   pre-print
Neural implicit functions have emerged as a powerful representation for surfaces in 3D.  ...  We propose to use iso-points as an explicit representation for a neural implicit function.  ...  The technique converts from an implicit neural surface to an explicit one via sampling iso-points, and goes back to the implicit representation via optimization.  ... 
arXiv:2012.06434v2 fatcat:guysfuk2b5cnjpljmb4hslezyu

DR3: Value-Based Deep Reinforcement Learning Requires Explicit Regularization [article]

Aviral Kumar, Rishabh Agarwal, Tengyu Ma, Aaron Courville, George Tucker, Sergey Levine
2021 arXiv   pre-print
effects of this implicit regularizer.  ...  feature representations.  ...  Finally, we inves- tigate the behavior of different implicit regularizers derived via two choices of M in Equation 4 and the corresponding explicit regularizers.  ... 
arXiv:2112.04716v1 fatcat:3jk67c5mpnathc263pzhyy47zi

Jacobian Norm for Unsupervised Source-Free Domain Adaptation [article]

Weikai Li, Meng Cao, Songcan Chen
2022 arXiv   pre-print
To combat this challenge, existing USFDAs turn to transfer knowledge by aligning the target feature to the latent distribution hidden in the source model.  ...  Then, following the theoretical insight, a general and model-smoothness-guided Jacobian norm (JN) regularizer is designed and imposed on the target domain to mitigate this dilemma.  ...  However, focusing on implicit alignment alone is often insufficient to address USFDA, since obtaining a desired alignment is usually difficult in the absence of source data, while such an implicit alignment  ... 
arXiv:2204.03467v1 fatcat:fnr6koayobfpnka5sp6z77cltq

Implicit Neural Deformation for Multi-View Face Reconstruction [article]

Moran Li, Haibin Huang, Yi Zheng, Mengtian Li, Nong Sang, Chongyang Ma
2021 arXiv   pre-print
, which learns to render on-surface points of the neural SDF to match the input images via self-supervised optimization.  ...  Unlike previous methods which are built upon 3D morphable models (3DMMs) with limited details, our method leverages an implicit representation to encode rich geometric features.  ...  Conclusions In this work, we present a novel method for 3D face reconstruction from multi-view images via implicit neural deformation.  ... 
arXiv:2112.02494v1 fatcat:gztidou42vfsxi2giol3jvelfa

Correspondence-Free Point Cloud Registration with SO(3)-Equivariant Implicit Shape Representations [article]

Minghan Zhu, Maani Ghaffari, Huei Peng
2021 arXiv   pre-print
We learn an embedding for each point cloud in a feature space that preserves the SO(3)-equivariance property, enabled by recent developments in equivariant neural networks.  ...  The proposed shape registration method achieves three major advantages through combining equivariant feature learning with implicit shape models.  ...  In a regular network, the feature matrix with feature dimension C corresponding to a set of N points is V ∈ R N ×C . In Vector Neuron networks, it is V ∈ R N ×C×3 .  ... 
arXiv:2107.10296v2 fatcat:i3g7kfg64zbgphxyng7wrtn7ty

Template NeRF: Towards Modeling Dense Shape Correspondences from Category-Specific Object Images [article]

Jianfei Guo, Zhiyuan Yang, Xi Lin, Qingfu Zhang
2021 arXiv   pre-print
Using periodic activation and feature-wise linear modulation (FiLM) conditioning, we introduce deep implicit templates on 3D data into the 3D-aware image synthesis pipeline NeRF.  ...  We present neural radiance fields (NeRF) with templates, dubbed Template-NeRF, for modeling appearance and geometry and generating dense shape correspondences simultaneously among objects of the same category  ...  Related Work Implicit Neural Representations and Rendering.  ... 
arXiv:2111.04237v1 fatcat:sulvlx4y6vfntnyi5pur5nhvxq

PVSeRF: Joint Pixel-, Voxel- and Surface-Aligned Radiance Field for Single-Image Novel View Synthesis [article]

Xianggang Yu, Jiapeng Tang, Yipeng Qin, Chenghong Li, Linchao Bao, Xiaoguang Han, Shuguang Cui
2022 arXiv   pre-print
Specifically, in addition to pixel-aligned features, we further constrain the radiance field learning to be conditioned on i) voxel-aligned features learned from a coarse volumetric grid and ii) fine surface-aligned  ...  Previous solutions, such as pixelNeRF, rely only on pixel-aligned features and suffer from feature ambiguity issues.  ...  The neural implicit functions [4, 28, 38, 41, 48] represent 3D surfaces by continuous functions defined in 3D space.  ... 
arXiv:2202.04879v1 fatcat:iacxkdhstfgrncxr4wxvux4zrm

Simultaneous Semantic Alignment Network for Heterogeneous Domain Adaptation [article]

Shuang Li, Binhui Xie, Jiashu Wu, Ying Zhao, Chi Harold Liu, Zhengming Ding
2020 arXiv   pre-print
Meanwhile, by leveraging target pseudo-labels, a robust triplet-centroid alignment mechanism is explicitly applied to align feature representations for each category.  ...  ., distinct domain distributions and difference in feature type or dimension.  ...  , target and source-target combined feature representations via refined target pseudo-labels.  ... 
arXiv:2008.01677v2 fatcat:my3hq5hbmnhqjkeoyo7p3dlh24

Local Implicit Grid Representations for 3D Scenes

Chiyu Jiang, Avneesh Sud, Ameesh Makadia, Jingwei Huang, Matthias NieBner, Thomas Funkhouser
2020 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
By localizing implicit functions in a grid, we are able to reconstruct entire scenes from points via optimization of the latent grid.  ...  (c) Reconstructing entire scenes with Local Implicit Grids Figure 1: We learn an embedding of parts from objects in ShapeNet [3] using a part autoencoder with an implicit decoder.  ...  To this end, we propose the Local Implicit Grid (LIG) representation, a regular grid of overlapping part-sized local regions, each encoded with an implicit feature vector.  ... 
doi:10.1109/cvpr42600.2020.00604 dblp:conf/cvpr/JiangSMHNF20 fatcat:geslahnq6rd67kuzw6ejoq7sme
« Previous Showing results 1 — 15 out of 11,939 results