A Dataset of Multi-Illumination Images in the Wild

Lukas Murmann*¹ Michael Gharbi¹ ² Miika Aittala¹ Fredo Durand¹
¹MIT CSAIL ²Adobe Research *lmurmann@mit.edu

Collections of images under a single, uncontrolled illumination have enabled the rapid advancement of core computer vision tasks like classification, detection, and segmentation. But even with modern learning techniques, many inverse problems involving lighting and material understanding remain too severely ill-posed to be solved with single-illumination datasets. To fill this gap, we introduce a new multi-illumination dataset of more than 1000 real scenes, each captured under 25 lighting conditions.

Download PDF from Arxiv
Browse the dataset
Get the code from Github

Data

We provide high-quality .jpg files as the default version of the dataset. In addition, we provide high-dynamic range images in .exr format. All data licensed under CC-BY .

There are two ways to access the data. First, users can download the entire test/train sets as .zip archives below. Second, scenes can also be downloaded on a per-scene basis using the Python SDK.

  • Download .exr HDR test set (1.5GB)
  • Download .exr HDR training set (50GB)
    •
    The entire set of full-resolution RAW files (over 2TB) is available for download. Please follow the instructions.

    Code

    Clone the SDK from Github 

    from matplotlib import pyplot as plt

import multilum

# load images for three scenes, 2 light directions
# access images as I[scene_index, light_index]
I = multilum.query_images(
    scenes=['main_experiment120',
            'kingston_dining10',
            'elm_revis_kitchen14'],
    dirs=[14,24]
)

# visualize in 2x3 grid
for i in range(3):
    plt.subplot(2,3,i+1)
    plt.imshow(I[i,0])

    plt.subplot(2,3,i+4)
    plt.imshow(I[i,1])

plt.show()

    Video

    Citation

    Bibtex 

    @inproceedings{murmann19,
  author={Lukas Murmann and Michael Gharbi and Miika Aittala and Fredo Durand},
  booktitle={2019 IEEE International Conference on Computer Vision (ICCV)},
  title={A Multi-Illumination Dataset of Indoor Object Appearance},
  year={2019},
  month={Oct},
}