2D-3D Match Dataset is a dataset that implements 2D-3D correspondence by leveraging several 3D datasets from RGB-D scans. The dataset uses data from SceneNN and 3DMatch. The training dataset consists of 110 RGB-D scans, 56 scenes from SceneNN and 54 scenes from 3DMatch. Given 3D points randomly sampled from a 3D point cloud, a set of 3D patches are extracted from different scan views. This dataset adopts a new method of learning local cross-domain descriptors for matching between 2D images and 3D point clouds. The dataset contains 1.4 million 2D-3D correspondences with different lighting conditions and settings obtained from public RGB-D scenes.

Citation

If you find our work useful for your research, please consider citing: @inproceedings{pham2020lcd, title = {{LCD}: {L}earned cross-domain descriptors for 2{D}-3{D} matching}, author = {Pham, Quang-Hieu and Uy, Mikaela Angelina and Hua, Binh-Son and Nguyen, Duc Thanh and Roig, Gemma and Yeung, Sai-Kit}, booktitle = {AAAI Conference on Artificial Intelligence}, year = 2020 } @inproceedings{hua2016scenenn, title = {{SceneNN}: {A} scene meshes dataset with a{NN}otations}, author = {Hua, Binh-Son, and Pham, Quang-Hieu and Nguyen, Duc Thanh and Tran, Minh-Khoi and Yu, Lap-Fai and Yeung, Sai-Kit}, booktitle = {International Conference on 3D Vision}, year = 2016 } @inproceedings{zeng20173dmatch, title = {{3DMatch}: {L}earning local geometric descriptors from {RGB}-{D} reconstructions}, author= {Zeng, Andy and Song, Shuran and Nie{\ss}ner, Matthias and Fisher, Matthew and Xiao, Jianxiong and Funkhouser, Thomas}, booktitle = {IEEE Conference on Computer Vision and Pattern Recognition}, year = 2017 }