Indoor Scene Reconstruction
Approximating shapes in images with low-complexity polygons [CVPR20] Floorplan generation from 3D point clouds: A space partitioning approach [ISPRS21] Plan2Scene: Converting Floorplans to 3D Scenes [CVPR21] MonteFloor: Extending MCTS for Reconstructing Accurate Large-Scale Floor Plans [ICCV21] HEAT: Holistic Edge Attention Transformer for Structured Reconstruction [CVPR22] Connecting the Dots: Floorplan Reconstruction Using Two-Level Queries [code]
Instant-NGP based methods
instant-ngp HashNeRF-pytorch This project is a pure PyTorch implementation of Instant-NGP. torch-ngp ngp_pl Instant-ngp (only NeRF) in pytorch+cuda trained with pytorch-lightning. 比上面的快，比原版的慢。 JNeRF JNeRF is an NeRF benchmark based on Jittor. JNeRF supports Instant-NGP capable of training NeRF in 5 seconds and achieves similar performance and speed to the paper. nerfstudio
OReX: Object Reconstruction from Planner Cross-sections Using Neural Fields paper
OReX reconstructs smooth 3D shapes (right) from input planar cross-sections (left).
看着有点意思 不知道有什么用
LoopDraw: a Loop-Based Autoregressive Model for Shape Synthesis and Editing paper
与上面类似
ActiveRMAP: Radiance Field for Active Mapping And Planning paper
NeRF做路径规划
ActiveNeRF: Learning where to See with Uncertainty Estimation [ECCV22]
NOPE-SAC: Neural One-Plane RANSAC for Sparse-View Planar 3D Reconstruction code
两张图片检测平面匹配以估计相机位姿
NeAF: Learning Neural Angle Fields for Point Normal Estimation paper
预测某个点法向与一些固定方向的query vectors之间的angle offset
GeoUDF: Surface Reconstruction from 3D Point Clouds via Geometry-guided Distance Representation code
NeuralUDF: Learning Unsigned Distance Fields for Multi-view Reconstruction of Surfaces with Arbitrary Topologies [ECCV22]
Weakly Supervised Semantic Segmentation for Large-Scale Point Cloud paper
弱监督点云上色做语义分割
NeuS2: Fast Learning of Neural Implicit Surfaces for Multi-view Reconstruction [Code (coming soon)]
Instant-NGP 加速了NeuS。做了动态场景的表示， For every subsequent frame, we predict its global transformation with respect to the previous frame and accumulate the transformation to convert it into the canonical space.
Real-Time Neural Light Field on Mobile Devices [page]
VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction [page]
与SparseNeuS对比，训练的数据集上效果好的，泛化性差不多，方法上一般
Flattening-Net: Deep Regular 2D Representation for 3D Point Cloud Analysis [Code(comingd)]
an unsupervised deep neural architecture called Flattening-Net to represent irregular 3D point clouds of arbitrary geometry and topology as a completely regular 2D point geometry image (PGI) structure