Intrinsic Image Evaluation

This page lists the different intrinsic image algorithms that we included in our evaluation, as well as their best parameters found from training. An example decomposition is included for each algorithm. See our publication [Bell et al. 2014] for more details.

The following is a quantitative comparison of our algorithm against several recent open-source algorithms, using mean weighted human disagreement rate (WHDR10%). All values were computed using leave-one-out cross validation.

Algorithm: bell2014_densecrf

Image

Reflectance

Shading

Citation:

Sean Bell, Kavita Bala, Noah Snavely. "Intrinsic Images in the Wild". ACM Transactions on Graphics (SIGGRAPH 2014). http://intrinsic.cs.cornell.edu.

Best training parameters:

  • abs reflectance weight: 0
  • abs shading gray point: 0.5
  • abs shading weight: 500.0
  • chromaticity weight: 0
  • kmeans intensity scale: 0.5
  • kmeans n clusters: 20
  • n iters: 25
  • pairwise intensity chromaticity: True
  • pairwise weight: 104
  • shading blur init method: none
  • shading blur sigma: 0.1
  • shading target norm: L2
  • shading target weight: 20000.0
  • split clusters: True
  • theta c: 0.025
  • theta l: 0.1
  • theta p: 0.1

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 21.0% (δ: 0.1)
  • Mean runtime: 214.4 s

View decompositions using these parameters.

Algorithm: zhao2012_nonlocal

Image

Reflectance

Shading

Citation:

Qi Zhao, Ping Tan, Qiang Dai, Li Shen, Enhua Wu, Stephen Lin. "A Closed-form Solution to Retinex with Non-local Texture Constraints". IEEE Transaction on Pattern Analysis and Machine Intelligence (TPAMI). http://www.ece.nus.edu.sg/stfpage/eletp/Papers/pami12_intrinsic.pdf.

Best training parameters:

  • chrom thresh: 0.001
  • gamma: False
  • texture patch distance: 0.0003
  • texture patch variance: 0.03

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 23.7% (δ: 0.1)
  • Mean runtime: 34.7 s

View decompositions using these parameters.

Algorithm: garces2012_clustering

Image

Reflectance

Shading

Citation:

Elena Garces, Adolfo Munoz, Jorge Lopez-Moreno, Diego Gutierrez. "Intrinsic Images by Clustering". Computer Graphics Forum (Eurographics Symposium on Rendering). http://www-sop.inria.fr/reves/Basilic/2012/GMLG12/.

Best training parameters:

  • km k: 8
  • remap gamma 2 2: False

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 25.7% (δ: 0.1)
  • Mean runtime: 5.1 s

View decompositions using these parameters.

Algorithm: grosse2009_grayscale_retinex

Image

Reflectance

Shading

Citation:

Roger Grosse, Micah K. Johnson, Edward H. Adelson, William T. Freeman. "Ground truth dataset and baseline evaluations for intrinsic image algorithms". Proceedings of the International Conference on Computer Vision (ICCV). http://www.cs.toronto.edu/~rgrosse/intrinsic/.

Best training parameters:

  • L1: True
  • threshold: 0.5

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 27.3% (δ: 0.1)
  • Mean runtime: 225.3 s

View decompositions using these parameters.

Algorithm: grosse2009_color_retinex

Image

Reflectance

Shading

Citation:

Roger Grosse, Micah K. Johnson, Edward H. Adelson, William T. Freeman. "Ground truth dataset and baseline evaluations for intrinsic image algorithms". Proceedings of the International Conference on Computer Vision (ICCV). http://www.cs.toronto.edu/~rgrosse/intrinsic/.

Best training parameters:

  • L1: True
  • threshold color: 0.7
  • threshold gray: 0.5

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 27.4% (δ: 0.1)
  • Mean runtime: 198.5 s

View decompositions using these parameters.

Algorithm: shen2011_optimization

Image

Reflectance

Shading

Citation:

Jianbing Shen, Xiaoshan Yang, Yunde Jia, Xuelong Li. "Intrinsic Images Using Optimization". Proceedings of Computer Vision and Pattern Recognition (CVPR). http://cs.bit.edu.cn/~shenjianbing/cvpr11.htm.

Best training parameters:

  • rho: 1.9
  • unmap srgb: False
  • wd: 3

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 32.4% (δ: 0.1)
  • Mean runtime: 297.1 s

View decompositions using these parameters.

Algorithm: baseline_reflectance

Image

Reflectance

Shading

Best training parameters: None

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 36.6% (δ: 0.1)
  • Mean runtime: 0.1 s

View decompositions using these parameters.

Algorithm: baseline_shading

Image

Reflectance

Shading

Best training parameters: None

Mean result across all 5230 photos:

  • Mean weighted human disagreement rate (WHDR): 51.6% (δ: 0.1)
  • Mean runtime: 0.1 s

View decompositions using these parameters.