TTD: Transparent Tableware Dataset

The Transparent Tableware Dataset (TTD) is a small dataset for image-based object pose estimation (mono or stereo). The dataset focuses on the localization of transparent objects. It contains 600 rectified stereo images of scenes of a table setting with some objects placed in the scene, including 5 transparent tableware objects. The table is set with 4 different tablecloths, and the images are divided accordingly (150 images per type of tablecloth). For each transparent object, the CAD model, pose annotations and segmentation masks are provided. Sometimes the transparent objects are partially or fully occluded since all objects are present in the scene at the same time.Note: Please refer to the ttd_documentation.pdf file for a complete documentation and, if you want to quickly use the dataset to train and test your method, skip directly to Sec. Default Benchmarks.Citation If you use this dataset in your research, please cite the following paper (the dataset was published together with this paper): I. Donadi and A. Pretto, "KVN: Keypoints Voting Network with Differentiable RANSAC for Stereo Pose Estimation," in IEEE Robotics and Automation Letters, vol. 9, no. 4, pp. 3498-3505, April 2024, doi: 10.1109/LRA.2024.3367508.BibTeX entry:@article{dp_RA-L2024, title={{KVN}: {K}eypoints Voting Network with Differentiable {RANSAC} for Stereo Pose Estimation}, author={Donadi, Ivano and Pretto, Alberto}, journal={IEEE Robotics and Automation Letters}, year={2024}, volume={9}, number={4}, pages={3498-3505}, doi={10.1109/LRA.2024.3367508}}