ARAP: The "As R[ealistic] As Possible" dataset for imaging applications.


This page offers ground truth decompositions and other quantities (depth, segmentations...) from realistic synthetic 3D renderings. If you are interested in this dataset and use it in your publications, please acknowledge the following state-of-the-art report :

@article{BKPB17,
author = {Bonneel, Nicolas and Kovacs, Balazs and Paris, Sylvain and Bala, Kavita},
title = {{Intrinsic Decompositions for Image Editing}},
journal = {Computer Graphics Forum (Eurographics State of The Art Report)},
year = {2017},
}

The original dataset used for benchmarking intrinsic decomposition algorithms can be found here.

Click on the scene thumbnail for related data.

Interiors
Animals
Objects
Outdoor
Humans


Clicking on the image will bring you to a page containing several of the following data:
- Full intrinsic decomposition (Reflectance and Shading layers)
- Additional photometric data (Irradiance and Direct shadows)
- Geometric data (Depth, Normals and Positions)
- Segmentation data (by objects and by materials, and an alpha mask)
Images have been rendered with LuxRender. To date, LuxRender offers the most Arbitrary Output Variables (AOV) options.


Limitations
- Intrinsic decomposition not well defined for specular or transparent objects.
- Intrinsic decomposition not well defined when there is blur (filtering, depth of field etc.).
- All of the other layers consider transparent objects as opaque.
- Artists sometimes bake shading into textures, which is then considered as reflectance.


Special thank to Jene Verwiebe for his help on #luxrender.