2D-3D-Matching

Case 1 : Papers for 2D-3D correspondences....

Year Paper link Code Notes
[ECCV2020] Learning 2D–3D Correspondences To Solve The Blind Perspective-n-Point Problem PDF1 PDF2 Github Youtube Output is the prioritized matches & estimated camera pose
[Arxiv, April2019] 2D3D-MatchNet: Learning to Match Keypoints Across 2D Image and 3D Point Cloud PDF github Output is the prioritized matches
[CVPR2018] Learning Less is More – 6D Camera Localization via 3D Surface Regression PDF github Output is the estimated camera pose
[ICCV2019] Expert Sample Consensus Applied to Camera Re-Localization (ESAC) PDF github Output is the estimated camera pose. It extended LessisMore(CVPR2018) for large scale dataset.
[CVPR2020] KFNet: Learning Temporal Camera Relocalization using Kalman Filtering PDF github Output is the estimated camera pose. SCoordNet is the 1-short retrieval version of KFNet.

Note: SCoordNet outperforms the state-of-the-art structure-based methods DSAC++ [6] and ESAC [7], yet with fewer parameters (24M vs. 210M vs. 28M, respectively).

Case 2 : Papers for Localization where 2D images is given to match 3D SfM....

Year Paper link Code Notes
ICCV 2019 Cascaded Parallel Filtering for Memory-Efficient Image-Based Localization PDF github SIFT is used in SfM database
CVPR 2019 HF-Net: Robust Hierarchical Localization at Large Scale PDF github SIFT/SuperPoints is used in SfM database

Case 3 : 3D point features....

Year Paper link Code Extra.
ECCV 2020 DH3D: Deep Hierarchical 3D Descriptors for Robust Large-Scale 6DoF Relocalization PDF github https://vision.in.tum.de/research/vslam/dh3d
ECCV2020 3D-CVF: Generating Joint Camera and LiDAR Features Using Cross-View Spatial Feature Fusion for 3D Object Detection PDF