Note the paper describing this dataset is available here along with the data files. When publishing results based on this database, please cite it as:
Xiangpeng Hao and Brian Funt, “A Multi-illuminant Synthetic Image Test Set,” Color Research and Application (in press), 2020.
Note that the specular component images are quite noisy since they are based on ray tracing using 800 random rays. This generally should not matter, but it could influence the results of any tests based on the specular component alone. The noise is most noticeable at sharp edges and at pixels with a very high specular component.
To download all images (~500GB) at once use the wget command:
*If you are looking for pixel-wise SPD*: the dataset contains three models, i.e. bathroom, butterfly, livingroom; each folder (e.g. livingroom) refers to a 3D model and contains six different light settings, each light setting (e.g. livingroom_exr_01) have three sub-folders, i.e. `combined`, `diffuse`, `glossy`.
`diffuse` refers to the diffuse component, `glossy` refers to the glossy component, and `combined` is the pixel-wise sum of previous two components(more details in the paper).
Spectral reflectances are in the `diffuse` folder with naming `_direct.exr`, for example, `livingroom/livingroom_exr_01/diffuse/vp_0_left_diffuse_direct.exr`.
*If you're looking for RGB images*: checkout the folders with `_rgb` postfix, e.g. `livingroom_rgb`, it contains the scene's depth_normal maps (`depth_normal`), spectral reflectances in RGB (`groundtruth_rgb`) and images with different light settings(e.g. livingroom_rgb/livingroom_exr_01`).
The RBG folders have the same directory structure as SPD folders except that every image in the RGB folder can be directly displayed in an EXR viewer.
Every RGB image is generated from the corresponding SPD image using the 2º-observer color matching functions.