Pop-Out Motion [Kor]
Lee et al. / Pop-Out Motion - 3D-Aware Image Deformation via Learning the Shape Laplacian / CVPR 2022
ģė
ķģøģ. ė³ø ķ¬ģ¤ķ
ģģė ģ¬ķ“ CVPRģ ė°ķė Pop-Out Motionģ“ė¼ė ė
¼ė¬øģ ģź°ėė¦¬ź³ ģ ķ©ėė¤. ģģ°ģ¤ė¬ģ“ 3D-Aware Image Deformationģ ģķ ķģµ źø°ė°ģ ķģ“ķė¼ģøģ ģ ģķ ė
¼ė¬øģ“ė©°, 3D Vision
, Shape Deformation
, 2D-to-3D Reconstruction
ė±ģ ķ¤ģėģ ź“ģ¬ģ“ ģģ¼ģ ė¶ė¤ģ“ė¼ė©“ ė
¼ė¬ø 본문 ė° ķė”ģ ķø ķģ“ģ§ė„¼ 구경ķ“주ģė©“ ź°ģ¬ķź² ģµėė¤. ķ“ė¹ ė
¼ė¬øģ ģ ź° 1ģ ģė” ģ°øģ¬ķģģ¼ė©°, KAIST ģ ģ°ķė¶ģ ģ±ėÆ¼ķźµģėź³¼ ź¹ķź· źµģėź»ģ ģ§ėķ“주ģ
Øģµėė¤. (ģ¢ģ ģ°źµ¬ ģ§ė넼 ķ“주ģ ė źµģėź» ź°ģ¬ė립ėė¤.)
1. Problem Definition
ė³ø ė
¼ė¬øģ 3D-Aware Image Deformation
ģ“ė¼ė 문ģ 넼 ķ“ź²°ķź³ ģ ķ©ėė¤. ģ¬ģ©ģź° ģ“ėÆøģ§ ė“ģ ź°ģ²“ ėŖØģģ ģģ°ģ¤ė½ź² ė³ķķė ź²ģ ź°ė„ķź² ķė, 2D ģģģ ķ¼ģ¬ģ²“ź° ė§ģ¹ 3D ź³µź°ģ 씓ģ¬ķė ź²ź³¼ ź°ģ“ ė³ķķ ģ ģėė” ķė ź²ģ“ ėŖ©ķģ
ėė¤. ģ“ ė ģ§ź“ģ ģø ģ“ėÆøģ§ ģģ ģ ģķģ¬ ģ¬ģ©ģź° ķ¤ķ¬ģøķø ė±ģ Deformation Handle
ģ 매ź°ģ²“ė”ģ ģ¬ģ©ķ ģ ģėė” ķ©ėė¤. ģėģ ź·øė¦¼ ģģ넼 볓ģė©“, ģ¬ģ©ģź° ģ“미ģ§ģ ķ¤ķ¬ģøķøė¤ (ķėģ ģ ķģ) ģ ģ§ģ ķź³ ź·ø ģ¤ ķė넼 ģ ķķģ¬ ģģ§ģ¼ ź²½ģ° (빨ź°ģ ķģ“ķ ķģ), 그림 ė“ģ ģ¬ė ź°ģ²“ ėŖØģģ“ ź·øģ ė§ģ¶ģ“ ģģ°ģ¤ė½ź² ė³ķėė ź²ģ ė³¼ ģ ģģµėė¤. ģ“ ė ķģ“ ėŖøķµ ė¶ė¶ ģģ ģģ¹ķź² ėź±°ė, ķ ė°ģ“ ė¤ė„ø ė° ė¤ė” ź°ė ¤ģ§ė ė±ģ 3D ź³µź°ģ ėķ ģ“ķ“넼 źø°ė°ģ¼ė” ķ ģģ ė³ķģ“ ģ¼ģ“ėź² ė©ėė¤. ģ“ė¬ķ 3D-Aware Image Deformation
źø°ė„ģ ģøķ°ė ķ°ėø ģģ ķøģ§ ģ“ķ리ģ¼ģ“ģ
ė±ģ ģ ģ©ķź² ģ°ģ¼ ģ ģģµėė¤.

2. Motivation
Related work
기씓ģė 3D ź³µź°ģ ėķ ģ“ķ“넼 źø°ė°ģ¼ė” ģģ ķøģ§ģ ź°ė„ķź² ķ źø°ė²ė¤ģ“ ė§ģ“ ģ°źµ¬ėģ“ ģģ§ė§, źøė”ė²ķ Scene ģ 볓 (ģ. ė·°ķ¬ģøķø, 칓ė©ė¼ ķė¼ėÆøķ°, ģ”°ėŖ ) ė ź¹ģ“ ģ 볓넼 ģģ ķė ź²ģ ģ ķėģ“ ģģģµėė¤. Human Pose Transfer ģŖ½ģ ģ°źµ¬ė¤ģ ģģ ģ ģ¬ėģ ģģøė„¼ ė³ķķė ź²ģ ź°ė„ķź² ķģ§ė§, ģ¬ėģ“ ģė ė¤ė„ø ģ¢ ė„ (ģ. ė§ķ ģŗė¦ķ°) ģ ģģ ģ ź°ģ²“ģ ėķ“ģė ėģķģ§ ģėė¤ė ķź³ģ ģ“ ģģģµėė¤. 3D ėŖØėø źø°ė° ė³ķ źø°ė²ė¤ģ ģģ ģ ź°ģ²“ ģ¢ ė„ģ źµķėģ§ ģź³ ėģķė¤ė ģ„ģ ģ“ ģģ§ė§, ģ ė „ ģģģ ėģėė ģ ķķ 3D ėŖØėøģ ķģė” ķė¤ė ėØģ ģ“ ģ”“ģ¬ķ©ėė¤. ģ“ė¬ķ ķź³ģ ė¤ģ ź°ģ ķźø° ģķģ¬ ģ ķ¬ ģ°źµ¬ģģė ź°ģ²“ ģ¢ ė„ģ źµķėģ§ ģź³ ģµėķ ģģ ė”ź² ģģ ė³ķģ“ ź°ė„ķ ķė ģģķ¬ė„¼ ź³ ģķė ź²ģ ėŖ©ķė” ķģģµėė¤.
Idea
ź°ģ²“ ģ¢ ė„ģ źµķėģ§ ģź³ ģµėķ ģģ ė”ź² ģģ ė³ķģ“ ź°ė„ķź² ķźø° ģķģ¬ ģ ė „ ģģģ¼ė”ė¶ķ° ė³µģė 3D Shapeģ ėķ“ Handle-Based Deformation Weight [1] ģ źø°ė°ģ¼ė” ģģ ė³ķģ ėŖØėøė§ķ©ėė¤. (1) Tetrahedral Mesh ķķģ 3D Shape ė° (2) ģ¬ģ©ģź° ģ§ģ ķ Deformation Handle ģ“ ģ£¼ģ“ģ”ģ ė, Handle-Based Deformationģ ė¤ģź³¼ ź°ģ“ ėŖØėøė§ė©ėė¤:
ģ ģģģģ ģ ė ģ ė „ Meshģ ė²ģ§ø Vertexģ ėķ ė³ķ ģ ė° ė³ķ ķ ģģ¹, ė Vertex ģ Handle ģ ėģėė Deformation Weight, ė ģ¬ģ©ģź° Handle ģ ź°ķė Affine Transformation ķė ¬ģ ģ미ķ©ėė¤.
ģ“ ė ģ¬ģ©ķė Handle-Based Deformation Weight [1] ģ ė¤ģź³¼ ź°ģ ģģģ ķµķ“ ź³ģ°ė©ėė¤:
ģ ģģģģ ź° Deformation Handleģ ėķ Deformation Weights ė Deformation Energy ģ ėķ Constrained Optimization 문ģ ģ ķ“ė”ģ ģ ģė©ėė¤.
ķ“ė¹ Deformation Energy ė ģ ė „ Meshģ Shape Laplacianģ ģ“ģ©ķģ¬ ģ ģėėė°, 2D-to-3D Reconstructionģ ķµķ“ ė³µģė Meshė”ė¶ķ°ė ė¶ģ ķķ Shape Laplacianģ“ ź³ģ°ėė¤ė ė¬øģ ź° ģģµėė¤. Shape Laplacianģ Mesh Topology (ģ¦, Mesh Vertex ź°ģ Edgeė”ģ ķķė ģ°ź²° ź“ź³) 넼 źø°ė°ģ¼ė” ķģ¬ ģ ģėėė°, 2D ģģģ¼ė”ė¶ķ° ģ ķķ Mesh Topology ģ 볓넼 ė³µģķ ģ ģė Topology-Aware Mesh Reconstructionģ ģ¬ė¬ ģ“ė ¤ģė¤ ė문ģ ģģ§ ķė¦¬ģ§ ģģ 문ģ ė” ėØģģģµėė¤. ė°ė¼ģ ģ ķ¬ģ ķµģ¬ ģģ“ėģ“ė 2Dė”ė¶ķ° ė³µģė 3D Shapeģ ėķ Shape Laplacian ģ 볓넼 ķģµ źø°ė°ģ źø°ė²ģ ķµķ“ ģ ķķź² ģģø”ķ ķ, ģ“넼 Handle-Based Deformation Weight ź³ģ°ģ ģ“ģ©ķė ź²ģ ėė¤.
3. Method
ģģ ģøźøėė øėÆģ“, ģ ķ¬ė 3D-Aware Image Deformationģ ėŖØėøė§ķźø° ģķ ķģµ źø°ė°ģ źø°ė²ģ ģ ģķ©ėė¤. ģ°ģ ģ ė „ ģģģ ėķģ¬ 3D Reconstruction Method (PIFu [2]) 넼 ģ ģ©ķØģ¼ė”ģØ ģģ ģ ź°ģ²“ģ ėģķė 3D Point Cloud넼 ģģø”ķ©ėė¤. (ģ ķ¬ė Mesh Edge ģ ė³“ź° ģ¬ģ©ėė Shape Laplacian ź³ģ°ģ ķģµ źø°ė°ģ źø°ė²ģ¼ė” ėģ²“ķ ź²ģ“źø° ė문ģ, Meshź° ģė Point Cloud ķķģ Shapeģ ģ¬ģ©ķ©ėė¤.) ė¤ģģ ė³µģė 3D Point Cloudģ ėķ Shape Laplacianģ ģøģ¬ķź² ģ¤ź³ė ė“ė“ė·ģ ģ“ģ©ķģ¬ ģģø”ķ©ėė¤. ģ“ė ź² ģģø”ė Shape Laplacianģ ģ“ģ©ķģ¬ ģ¬ģ©ģź° ģģė” ģ§ģ ķ Deformation Handleģ ėķ Handle-Based Deformation Weight [1]ģ ź³ģ°ķź³ , ģ“넼 ķµķ“ ėŖØėøė§ ė 3D Deformationģ ė¤ģ 2D Image Planeģ ķ¬ģ¬ķØģ¼ė”ģØ 3D-Aware Image Deformationģ ź°ė„ķź² ķ©ėė¤.
ģ§źøė¶ķ°ė ģ ķ¬ģ ķµģ¬ ģģ“ėģ“ģø Point Cloudė”ė¶ķ° Shape Laplacianģ ģģø”ķė ė¤ķøģķ¬ģ ėķģ¬ ģģøķź² ģź°ėė¦¬ź² ģµėė¤. Shape Laplacianģ źµ¬ģ± ģģģø Cotangent Laplacian Matrix ģ Inverse Mass Matrix 넼 ė°ė” ģģø”ķėė” ė¤ķøģķ¬ė„¼ 구ģ±ķ ķ, ź° ģ 볓ģ ėķ ģ§ģ ģ ģø Superivsionģ ķµķģ¬ ė¤ķøģķ¬ė„¼ ķģµģķµėė¤. ģėģ ź·øė¦¼ģ 볓ģė©“ ģ ģ ģėÆģ“, ģ ģ ķė ģģķ¬ė ķ¬ź² ģø ź°ģ§ģ ėŖØė - (1) Feature Extraction Module, (2) Cotangent Laplacian Prediction Module, (3) Inverse Mass Prediction Module - ė” źµ¬ģ±ėģ“ģģµėė¤.

Feature Extraction Moduleģ ģ ė „ 2D ģ“미ģ§ė”ė¶ķ° ė³µģė 3D Point Cloud 넼 ģ ė „ģ¼ė” ė°ģ Point Cloud Feature 넼 ģģ±ķ©ėė¤. ģ“ ė ģ ģ ėģėė Per-Point Feature넼 ģ미ķ©ėė¤. ėŖØėģ źµ¬ģ”°ė”ė Point Transformer [3] 넼 ķģ©ķģģµėė¤.
Cotangent Laplacian Prediction Moduleģ 3D Point Cloud ģ Point Cloud Feature 넼 ģ
ė „ģ¼ė” ė°ģ ģ ėķ Cotangent Laplacian Matrix 넼 ģģø”ķ©ėė¤. Cotangent Laplacianģ ģ ģģ ė°ė¼ ģ Symmetricķź³ ė§¤ģ° Sparseķ ķ¹ģ±ģ ź°ģ§ź³ ģėė°, ģ ģ¬ģ“ģ Edge ģ°ź²° ź“ź³ź° ģģ“ģ¼ ģ“ 0ģ“ ģė ź°ģ¼ė” ģ ģėźø° ė문ģ
ėė¤. ģ ķ¬ė Point Cloud ė“ģ ź° Point Pair (, ) 넼 ģ
ė „ģ¼ė” ė°ģ ģ“ģ ėģėė Laplacian Matrixģ Element () 넼 ė³ė ¬ģ ģ¼ė” ģģø”ķė źµ¬ģ”°ė„¼ ģ·Øķėė°, Euclidean Distanceź° ėØ¼ Point Pair ė¼ė¦¬ė ģ°ź²° ź“ź³ź° ģģ ķė„ ģ“ ģ źø° ė문ģ ģ“ė¤ģ 1ģ°Øģ ģ¼ė” ź±øė¬ģ£¼ė ģķ ģ ķ©ėė¤. ė
¼ė¬øģģ KNN-Based Point Pair Sampling (KPS)
ģ¼ė” ģ§ģ¹ķė ė¶ė¶ģøė°, ź° ķ¬ģøķøė¤ģ ėķģ¬ ź°ģ ź°ź¹ģ“ ģ ė¤ģ ėķ“ģė§ Point Pair넼 구ģ±ķė źø°ė²ģ
ėė¤. ģ“ė¬ķ Sampling źø°ė²ģ ģ°ģ§ ģģ ź²½ģ° Imbalanced Regression Problemģ“ ģ¼ģ“ė ė¤ķøģķ¬ ķģµģ“ ģ ėģ§ ģė ķģģ“ ģģģµėė¤.
ė¤ģģ KNN-Based Point Pair Sampling (KPS)
ģ ķµķ“ ģ ķė ź° Point Pair Candidate (, ) ģ ėķģ¬ Symmetric Feature Aggregation
ģ ģķķ“ģ¤ėė¤:
ģ ģģģģ ė° ė”ė Symmetric Functionģ ģ¬ģ©ķėė°, ģ“ė ėģ¤ģ ģģø”ė Cotangent Laplacian Matrixģ Symmetry넼 볓ģ„ķźø° ģķØģ ėė¤. ķ“ė¹ ķØģė ź°ź° Absolute Differenceģ Element-Wise Multiplicationģ¼ė” 구ķėģģµėė¤. ģ“ė ź² ģģ±ė Point Pair Feature ģ ėģėė Cotangent Laplacian Element ė ė¤ģź³¼ ź°ģ“ ģģø”ė©ėė¤:
ģ Real-Valued Scalar넼 ģ¶ė „ķė ķØģģ“ė©° ė ģ“ Non-Zero ź°ģ¼ģ§ģ ėķ ķė„ ģ ėŖØėøė§ķė Weight ģ¶ė „ ķØģģ ėė¤. ė ķØģė MLPė” źµ¬ķėģģ¼ė©°, ģµģ¢ ź°ģ ė ģ¶ė „ ź°ģ ź³±ģ¼ė”ģ ķķė©ėė¤.
Inverse Mass Prediction Moduleģ 3D Point Cloud ģ Point Cloud Feature 넼 ģ ė „ģ¼ė” ė°ģ ģ ėķ Inverse Mass Matrix 넼 ģģø”ķ©ėė¤. Inverse Massģ ģ ģģ ė°ė¼ ģ Diagonal ķė©°, ė²ģ§ø Digonal Elementė ģ Volumeź³¼ ź“ź³ė ģ 볓넼 ė“ź³ ģģµėė¤. ė°ė¼ģ ė“ģ ź° ķ¬ģøķø ģ ėģėė Per-Point Feature 넼 Concatenate ģģ¼ģ¤ ķ MLPģ ķµź³¼ģķ¤ė ė°©ģģ ķµķ“ Inverse Mass Matrix ė“ģ Element넼 ģģø”ķ©ėė¤.
ė³ø Shape Laplacian ģģø” ė¤ķøģķ¬ė , , ģģø” ź°ģ ėķ L1-Loss źø°ė°ģ Ground Truth Supervisionģ ķµķ“ ķģµė©ėė¤. ģģøķ Loss ź³ģ° ģ 볓ė ė ¼ė¬ø 본문ģ ģ°øģ”°ķ“주ģė©“ ź°ģ¬ķź² ģµėė¤.
4. Experiment & Result
ģ ģķ 3D-Aware Image Deformation źø°ė²ģ ķØź³¼ģ±ģ ź²ģ¦ķźø° ģķģ¬ ķ¬ź² ė ģ¢ ė„ģ ģ¤ķģ ģ§ķķģģµėė¤. 첫 ė²ģ§øė”ė ģ ķ¬ź° ėŖØėøė§ķ Deformationģ ķ리ķ°ė„¼ ģ ėģ ģ¼ė” ķź°ķźø° ģķ“ 3D Point Cloud Deformation ģ¤ķģ ģ§ķķģģµėė¤. ė ė²ģ§øė”ė ģ ķ¬ģ ėŖ©ķ źø°ė„ģø 3D-Aware Image Deformation 결과넼 ķģøķźø° ģķ ģ ģ±ģ ķź°ė„¼ ģ§ķķģģµėė¤. ėģ± ė¤ģķ ģ¤ķ ź²°ź³¼ (ģ. Partial Point Cloud Deformation, Ablation Study) ė ė ¼ė¬ø 본문ģģ ķģøķ“주ģė©“ ź°ģ¬ķź² ģµėė¤.
Experimental setup
Dataset
DFAUST [4]: ģ ėģ ķź°ģ ģ¬ģ©ė 3D Human Point Cloud ė°ģ“ķ°ģ ģ ėė¤.
RenderPeople [5], Mixamo [6]: ģ ģ±ģ ķź°ģ ģ¬ģ©ė 3D Human [5] ė° 3D Character [6] Datasetģ ėė¤. ģ ķ¬ģ ėŖ©ģ ģ Image Deformationģ ź²°ź³¼ė„¼ ķģøķė ź²ģ“ėÆė”, ķ“ė¹ 3D Modelė¤ģ ė ėė§ķģ¬ ģģ±ķ ģģė¤ģ ģ¤ķģ ģ¬ģ©ķģģµėė¤.
Baselines
ģ ķ¬ģ ķµģ¬ ģģ“ėģ“ė Mesh Reconstruction ź²°ź³¼ė”ė¶ķ° ė¶ģ ķķ Shape Laplacianģ“ ź³ģ°ėėÆė” ķ“ė¹ ģ 볓넼 ķģµ źø°ė°ģ źø°ė²ģ ķµķ“ ė³“ė¤ ģ ķķź² ģģø”ķģė ź²ģ“ģģµėė¤. ė°ė¼ģ, Mesh Reconstruction źø°ė²ģ ģ¬ģ©ķģ¬ Shape Laplacianģ ģ»ģ ķ Deformation Weightģ ź³ģ°ķė ģķ©ģ ė² ģ“ģ¤ė¼ģøģ¼ė” ģ¤ģ ķģģµėė¤. ģ ķ¬ ģ¤ķģģ ź³ ė ¤ė Mesh Reconstruction źø°ė²ė¤ģ ė¤ģź³¼ ź°ģµėė¤:
Screened Poisson Surface Reconstruction (PSR) [7],
Algebraic Point Set Surfaces (APSS) [8],
Ball-Pivoting Algorithm (BPA) [9],
DeepSDF [10],
Deep Geometric Prior (DGP) [11],
Meshing Point Clouds with IntrinsicExtrinsic Ratio (MIER) [12].
ėķ, 기씓ģ Point Cloud Laplacian źø°ė²ģ ģ“ģ©ķģ¬ ģ ė „ Point Cloudė”ė¶ķ° Shape Laplacianģ ź·¼ģ¬ ź°ģ ė°ė” ź³ģ°ķė źø°ė²ė¤ė ź³ ė ¤ķģģµėė¤:
PCD Laplace (PCDLap) [13],
Nonmanifold Laplacians (NMLap) [14].
Training Setup
ź° ė°ģ“ķ°ė³ė” ģ¤ķģ ģ¬ģ©ķ ģøķ ģ“ ė¤ė„“ėÆė”, ģģøķ ģ¬ķģ ė ¼ė¬ø 본문 ė° Supplementary넼 ģ°øź³ ķ“주ģė©“ ź°ģ¬ķź² ģµėė¤.
Evaluation Metric
ģ ķ¬ģ ģ ėģ ķź°ģė ė¤ģź³¼ ź°ģ ė©ķøė¦ģ“ ģ¬ģ©ėģģµėė¤:
ģģø” ė° ģ ėµ Deformation Weights ź°ģ L1 Distance (Weight L1),
ģģø” ė° ģ ėµ Deformed Shape ź°ģ Chamfer Distance (Shape CD),
ģģø” ė° ģ ėµ Deformed Shape ź°ģ Hausdorff Distance (Shape HD).
Result
3D Point Cloud Deformation
ģėģ ķė DFAUST [4] ė°ģ“ķ°ģ ģ ėķ ģ ėģ ė¹źµ ķź° 결과넼 ėķėø ź²ģ ėė¤. ģ ķ¬ź° ģ ģķ źø°ė²ģ“ ė¤ė„ø Mesh Reconstruction ė² ģ“ģ¤ė¼ģø źø°ė²ė¤ģ ģ¬ģ©ķģ ė ė³“ė¤ ė ėģ Shape Deformation 결과넼 볓ģ“ė ź²ģ ģ ģ ģģµėė¤.

ģģ ź²°ź³¼ģ ėķ ģ ģ±ģ ź²°ź³¼ (ģė ź·øė¦¼) ėķ ģ ķ¬ źø°ė²ģ“ ėģ± ģģ°ģ¤ė¬ģ“ Shape Deformationģ ėŖØėøė§ķ ģ ģģģ 볓ģ¬ģ¤ėė¤.

3D-Aware Image Deformation
ė³ø ėģģģ ģ ķ¬ģ 3D-Aware Image Deformation źø°ė²ģ ģ“ģ©ķ“ģ ģģ±ķ ėŖØģ ėģģģ ėė¤. Mesh Reconstruction ė² ģ“ģ¤ė¼ģø źø°ė²ė¤ė³“ė¤ ėģ± ģģ°ģ¤ė¬ģ“ Image Deformationģ ģģ±ķ ģ ģģģ 볓ģ¬ģ¤ėė¤.

Interactive Demoė 첓ķķ“ė³“ģźø°ė„¼ ė°ėėė¤. ģ¬ģ©ģź° ģ§ź“ģ ģø Deformation Handle (Keypoint)넼 ģ“ģ©ķģ¬ ģģģ ė³ķķ ģ ģģµėė¤.
5. Conclusion
ė³ø ģ°źµ¬ģģė Shape Laplacianģ ķģµķØģ¼ė”ģØ ė³“ė¤ ģģ°ģ¤ė¬ģ“ 3D-Aware Deformationģ ź°ė„ķź²ķė ķė ģģķ¬ė„¼ ģ ģķģģµėė¤. ģ ķ¬ź° ģźø°ė”ė ģ“ź° ė“ė“ė· źø°ė° źø°ė²ģ“ Shape Lapacian ģģø”ģ ķØź³¼ģ ģ¼ ģ ģģģ ģ²ģģ¼ė” ė³“ģø ģ°źµ¬ė¼ź³ ģź³ ģģµėė¤. ė³ø ķė ģģķ¬ė„¼ ė°ģ ģķ¤źø° ģķ ėģ± ė¤ģķ ģģ“ėģ“ź° ė§ģė°, źø°ķź° ėė¤ė©“ ķ“ė¹ ė°©ķ„ģ¼ė” ėģ± ģ°źµ¬ķ“ė³“ź³ ģ¶ģµėė¤.
Take-Home Message (ģ¤ėģ źµķ)
ģ ź° ź°ģøģ ģ¼ė” ģ“ ķė”ģ ķøė„¼ ķµķ“ ė°°ģ“ źµķģ "ėź¹ģ§ ķ¬źø°ķģ§ ģź³ ģ§ė ģ ź°ģ§ė©° ģ°źµ¬ 문ģ 넼 ķģ"ė ź²ģ ėė¤. ė³ø ķė ģģķ¬ ź°ė° ėØź³ģģ ģģķ Challengeė¤ģ“ ė§ģģź³ , ź·ø ź³¼ģ ģ¤ ģė ģ§ķķė ¤ė ģ°źµ¬ ė°©ķ„ģ¼ė”ė¶ķ° ķ¬ź² ė°ėģ“ ė§ė¬“리ė ė¶ė¶ė ģģµėė¤. ź·øėė ėģķė ģ루ģ ģ ģ°¾ģė“ź³ ė» ź¹ź² ķė”ģ ķøė„¼ ė§ė¬“리ķ ģ ģģ“ģ ź°ģøģ ģ¼ė”ė ė§¤ģ° źø°ģµģ ėØė ģ°źµ¬ ź²½ķģ“ ėģģµėė¤. ź·ø ź³¼ģ ģ¤ ķ° ėģź³¼ ģ”°ģøģ 주ģ ė ģ§ė źµģėź» ź¹ģ ź°ģ¬ė„¼ ė립ėė¤.
Author / Reviewer information
Author
ģ“ģ§ķ (Jihyun Lee)
KAIST CS
I am a first-year Ph.D. student in Computer Vision and Learning Lab at KAIST advised by Prof. Tae-Kyun Kim. I am also currently co-advised by Prof. Minhyuk Sung. My research interests lie in machine learning for 3D computer vision and graphics - especially on humans.
Reviewer
Korean name (English name): Affiliation / Contact information
Korean name (English name): Affiliation / Contact information
...
Reference & Additional materials
Citation of related work
Alec Jacobson, Ilya Baran, Jovan Popovic, and Olga Sorkine. Bounded biharmonic weights for real-time deformation. In SIGGRAPH, 2011.
Shunsuke Saito, Zeng Huang, Ryota Natsume, Shigeo Morishima, Angjoo Kanazawa, and Hao Li. PIFu: Pixel-aligned implicit function for high-resolution clothed human digitization. In ICCV, 2019.
Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In ICCV, 2021.
Federica Bogo, Javier Romero, Gerard Pons-Moll, and Michael J. Black. Dynamic FAUST: Registering human bodies in motion. In CVPR, 2017.
Renderpeople GmbH. RenderPeople. https://renderpeople.com/.
Adobe Systems Inc. Mixamo. https://www.mixamo.com.
Michael Kazhdan and Hugues Hoppe. Screened poisson surface reconstruction. ACM TOG, 2013.
Gael Guennebaud and Markus Gross. Algebraic point set surfaces. In SIGGRAPH, 2007.
Fausto Bernardini, Joshua Mittleman, Holly Rushmeier, Claudio Silva, and Gabriel Taubin. The ball-pivoting algorithm for surface reconstruction. IEEE TVCG, 1999.
Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. DeepSDF: Learning continuous signed distance functions for shape representation. In CVPR, 2019.
Francis Williams, Teseo Schneider, Claudio Silva, Denis Zorin, Joan Bruna, and Daniele Panozzo. Deep geometric prior for surface reconstruction. In CVPR, 2019.
Minghua Liu, Xiaoshuai Zhang, and Hao Su. Meshing point clouds with predicted intrinsic-extrinsic ratio guidance. In ECCV, 2020.
Mikhail Belkin, Jian Sun, and Yusu Wang. Constructing laplace operator from point clouds in rd. In Proc. Annu. ACM-SIAM Symp. Discrete Algorithms, pages 1031ā1040. SIAM, 2009.
Nicholas Sharp and Keenan Crane. A laplacian for nonmanifold triangle meshes. In SGP, 2020.
Last updated
Was this helpful?