metadata
license: mit
task_categories:
- image-to-image
doc3D
Doc3D is the first 3D dataset focused on document unwarping with realistic paper warping and renderings.
It contains 100k images with the following ground-truths:
- 3D Coordinates
- Depth
- UV
- Backward Mapping
- Albedo
- Normals
- Checkerboard
Useful links:
- More details of the data usage instructions are available in the GitHub repo: https://github.com/cvlab-stonybrook/doc3D-dataset
- Link to the training code: https://github.com/cvlab-stonybrook/DewarpNet
- Link to the data generation code: https://github.com/sagniklp/doc3D-renderer
Citation:
If you use the dataset, please consider citing our work-
@inproceedings{SagnikKeICCV2019,
Author = {Sagnik Das*, Ke Ma*, Zhixin Shu, Dimitris Samaras, Roy Shilkrot},
Booktitle = {Proceedings of International Conference on Computer Vision},
Title = {DewarpNet: Single-Image Document Unwarping With Stacked 3D and 2D Regression Networks},
Year = {2019}}
Acknowlegement:
- Bash scripts are adapted from epic-kitchens-download-scripts.
- Textures are obtained from:
- Yes! Magazine under Creative Commons Licence.
- CVF Open Access
- From books available under Project Gutenberg