Learning Intrinsic Image Decomposition from Watching the World

Zhengqi Li    Noah Snavely

Cornell University/Cornell Tech

In CVPR, 2018 (Spotlight Oral)


Single-view intrinsic image decomposition is a highly ill-posed problem, making learning from large amounts of data an attractive approach. However, it is difficult to collect ground truth training data at scale for intrinsic images. In this paper, we explore a different approach to learning intrinsic images: observing image sequences over time depicting the same scene under changing illumination, and learning single-view decompositions that are consistent with these changes. This approach allows us to learn without ground truth decompositions, and instead to exploit information available from multiple images. Our trained model can then be applied at test time to single views. We describe a new learning framework based on this idea, including new loss functions that can be efficiently evaluated over entire sequences. While prior learning-based intrinsic image methods achieve good performance on specific benchmarks, we show that our approach generalizes well to several diverse datasets, including MIT intrinsic images, Intrinsic Images in the Wild and Shading Annotations in the Wild.

paper thumbnail

[Paper] [Supplemental]

Zhengqi Li and Noah Snavely. "Learning Intrinsic Image Decomposition from Watching the World".

  title={Learning Intrinsic Image Decomposition from Watching the World},
  author={Zhengqi Li and Noah Snavely},
  booktitle={Computer Vision and Pattern Recognition (CVPR)},


The BigTime dataset includes > 200 timelapse image sequences we collected from Internet.

BigTime_v1 Datasets (tar.gz, 16 GB)

Pretrained Model

Pretrained model (265 MB)

Training/Evaluation Code



We thank Jingguang Zhou for his help with data collection and running evaluations. We also thank the anonymous reviewers for their valuable comments. This work was funded by the National Science Foundation through grant IIS-1149393, and by a grant from Schmidt Sciences.