Ground truth dataset and baseline evaluations for intrinsic image algorithms

By Roger Baker Grosse, Micah K. Johnson, Edward H. Adelson, William T. Freeman. IEEE International Conference on Computer Vision, 2009

508 citations recorded by the source corpus.

The intrinsic image decomposition aims to retrieve “intrinsic” properties of an image, such as shading and reflectance. To make it possible to quantitatively compare different approaches to this problem in realistic settings, we present a ground-truth dataset of intrinsic image decompositions for a variety of real-world objects. For each object, we separate an image of it into three components: Lambertian shading, reflectance, and specularities. We use our dataset to quantitatively compare several existing algorithms; we hope that this dataset will serve as a means for evaluating future work on intrinsic images.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/ff947445afc8fb16eb8cb8986d5a9bb6eada155a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Single Image Layer Separation Using Relative Smoothness (2014, background): “A discussion and summary concludes the paper in Section 5.”
Learning Large-Scale Automatic Image Colorization (2015, background): “has been extensively studied, but is small and shows isolated objects of quite limited material complexity [6].”
Intrinsic Decompositions for Image Editing (2017, unknown): “This often leads to sparse linear systems [GJAF09].”
Rendering synthetic objects into legacy photographs (2011, methodology): “Recent work by Grosse et al. demonstrates that the color variant of Retinex is state-of-the-art for single-image decomposition methods [Grosse et al. 2009].”
Intrinsic Scene Properties from a Single RGB-D Image (2013, methodology): “We compare against two intrinsic image algorithms: Retinex [12, 15] and Gehler et al. [11].”
3DNN: 3D Nearest Neighbor (2014, methodology): “Their method uses intrinsic image decomposition to determine the albedo, direct illumination and indirect illumination of a scene by building upon Grosse et al. (2009) properties of the scene, Karsch et al. (2011) allow users to position new 3D objects into the scene which are photore-alistically…”
Shape, albedo, and illumination from a single image of an unknown object (2012, methodology): “This is the insight that underlies the Retinex algorithm [11, 16, 19], and informs more recent intrinsic images work [2, 26, 27, 28].”
Self-Supervised Intrinsic Image Decomposition (2017, methodology): “For example, the MIT Intrinsic Images dataset [Grosse et al., 2009], containing twenty real-world images, is not large enough for the unsupervised learning to affect the representations of our model. In the absence of any unsupervised training, our model is similar to that of Shi et al. [2017] adapt”
CGIntrinsics: Better Intrinsic Image Decomposition through Physically-Based Rendering (2018, methodology): “MPI Sintel [34] 890 Animation non-PB spatial-varying full MIT Intrinsics [35] 110 Object Real single global full ShapeNet [2] 2M+ Object PB single global full IIW [5] 5230 Scene Real spatial-varying sparse SAW [6] 6677 Scene Real spatial-varying sparse CGINTRINSICS 20,000+ Scene PB spatial-varying f”
Ieee Transactions on Pattern Analysis and Machine Intelligence Shape, Illumination, and Reflectance from Shading (unknown): “This is the insight that underlies the Retinex algorithm [4], [9], [10], and informs more recent intrinsic images work [13], [14], [15].”
Portrait shadow manipulation (2020, methodology): “ng the shadow detection step (Ma et al. 2016). General techniques for inverse rendering (Ramamoorthi and Hanrahan 2001; Sengupta et al. 2018) and intrinsic image decomposition (Barron and Malik 2015; Grosse et al. 2009) should, in theory, be useful for shadow removal, as they provide shading and re…”
Color Constancy, Intrinsic Images, and Shape Estimation (2012, methodology): “The twenty objects in the MIT Intrinsic Images dataset [5], with the ground-truth shape and illumination provided by [1].”
Low-Light Image Enhancement via a Deep Hybrid Network (2019, background): “Retinex theory [34] assumes that the amount of light reaching observers can be decomposed into illumination and scene reflection [35]–[37].”
LIME: Low-Light Image Enhancement via Illumination Map Estimation (2017, methodology): “The model of our problem is similar with that of the intrinsic image decomposition [16], [17], [18], [19], which attempts (b) Reflectancee”