[Registration] Inverse depth loss in SurgicalGS, EndoGaussian, Deform3dgs
Inverse depth maps compress the dynamic range of depth values, reducing the disparity between the binocular and rendered depth maps
Registration
[Registration] How to recover depth maps from predicted dispairty & RGB image warping loss
Left image만 input으로 하여, left, right disparity를 예측하고, 예측된 두 disparity로부터 camera focal length와 두 카메라 간의 baseline distance로 계산하여 depth map을 복원할 수 있습니다.
[Registration] Depth hypothesis, Disparity hypothesis, Cost Volume
Cost volume의 channel 수는 disparity hypothesis의 수와 관련이 있습니다.
[Registration] Probability volume, depth map inference, per-pixel depth estimation, estimation confidence, depth hypotheses
Probability volume, depth map inference, per-pixel depth estimation, estimation confidence
[Registration] Large-scale scene reconstruction’s challenge
transient objects, highly unlikely to be collected in a single caputre under consistent conditions, variance in both scene geometry and appearance
[Registration] 3D reconstruction process
3D reconstruction pipeline: Extract 2D features, matche features across different images, jointly optimize a set of 3D points and camera poses to be consiste...
[Registration] SuperGlue, Predator, REGTR
Matching 2D image correspondences: SuperGlue, Predator, REGTR
[Registration] global feature-extracting strategy
global feature-extracting strategy
[Registration] Evaluation Metircs: RRE, RTE, RSE, RDE, RR(Success Ratio), ATE
Relative Rotation Error (RRE), Relative Translation Error (RTE), Relative Scale Error (RSE), Relative Depth Error (RDE), Absolute Translational Error (ATE)
[Registration] Kabsch-Umeyama algorithm
Kabsch-Umeyama algorithm
[Registration] Kernel Point Convolution (KPConv)
Kernel Point Convolution (KPConv) is used to obtain the downsampled voxel points
[Registration] Rule based feature matching vs Deep Learning based feature matching
규칙 기반의 특징 점 매칭은 결국 특징점의 디스크립터간의 관계를 파악하는 구조에서 그칩니다.
[Registration] Rigid point cloud registration
Rigid point cloud registration refers to the problem of finding the optimal rotation and translation parameters that align two point clouds.
[Registration] Point cloud registration is for large-scale 3D scene reconstruction
Point cloud registration is a fundamental problem for large-scale 3D scene scanning and reconstruction
[Registration] SuperPoint & MagicPoint
MagicPoint는 interest points 검출에 초점을 맞춘 모델입니다.
[Registration] Interest point detection, descriptor learning
Interest point detection은 이미지에서 특징이 되는 부분, 즉 keypoint를 찾는 과정입니다.
[Registration] Image substructures
Image substructures는 이미지 내에서 의미 있는 부분이나 영역을 나타내는 개념입니다.
[Registration] Homograpy Estimation & Homographic Adaptation
Homography Estimation과 Homographic Adaptation은 관련은 있지만 다른 개념입니다.
[Registration] Homographic Adaptation
Homographic adaptation은 다양한 호모그래피 변환을 적용한 이미지들에서 관심점을 검출하고 이를 원본 이미지로 역변환하여 pseudo ground truth를 생성하는 자기 지도 학습 기법입니다.
[Registration] Edge, Corner 정의
이미지에서 일반적으로 가장 중요하고 구분하기 쉬운 특징점들은 코너와 엣지입니다.
[Registration] keypoint, descriptors
Descriptors는 특징점(keypoint)의 고유한 특성을 수치화한 벡터입니다.
[Registration] Speeded Up Robust Features (SURF)
Speeded Up Robust Features (SURF)는 SIFT 알고리즘을 기반으로 사용
[Registration] Scale-Invariant Feature Transform (SIFT)
the Scale-Invariant Feature Transform, or SIFT, is still probably the most well-known traditional local feature descriptor in computer vision.
[Registration] LIFT: Learned Invariant Feature Transform
LIFT(Learned Invariant Feature Transform)는 딥러닝 기반의 특징점 추출 및 매칭 기법입니다.
[Registration] Harris corner detection
Harris 코너 검출 알고리즘은 이미지에서 코너를 추출 하는 알고리즘의 대표적인 예입니다.
[Registration] HLoc = SuperPoint (feature extractor) + SuperGlue (matcher)
HLoc = SuperPoint(as feature extractor) + SuperGlue(as feature matcher)
[Registration] Features from Accelerated Segment Test (FAST)
Features from Accelerated Segment Test (FAST)의 경우 실시간 특징점 추출을 고려하여 만들어진 알고리즘
[Registration] Superpoints, Repeatability, descriptor matching
downsample the input point clouds into superpoints and then match them through examining whether their local neighborhood (patch) overlaps.
[Registration] Repeatable keypoints, keypoint-free method, putative correspondences
Repeatable keypoints, putative correspondences
[Registration] Global Registration & Local Registration
Global & Local Registartion