Part Labels Database
Welcome to the Part Labels Database! This database contains labelings of 2927 face images into Hair/Skin/Background labels. The face images are a subset of the Labeled Faces in the Wild (LFW) funneled images. Each funneled image is first segmented into superpixels and then the superpixels are manually labeled as one of the Hair/Skin/Background classes. Above, we show a sample LFW funneled image, the same image with superpixels superimposed, and the final labeled image.
We provide links to the data and training/test/validation sets used for evaluation. More details can be found in the paper below.
Download the database:
We provide links to the data and training/test/validation sets used for evaluation. More details can be found in the paper below.
Note that the funneled image are available here.
- Superpixels as gzipped tar file (23MB, md5sum 5d8a73979f5675b3e7ae147c7443b15e)
- Superpixel Features as gzipped tar file (91MB, md5sum d3c57a005a6369002e253f3458beef3a)
- Ground Truth Labels as gzipped tar file (1MB, md5sum 02196a0ab9e62bec5e8a9e724005a519)
- Ground Truth Images as gzipped tar file (6MB, md5sum 3e7e26e801c3081d651c8c2ef3c45cfc)
Training, Validation, Testing:
We use the following files for training/validation/testing. There are 1500 images in training, 500 used in validation, and 927 used for testing.
[Train]
[Validation]
[Test]
Code:
The following code is used to generate the features.
[gloc_features.zip] (md5sum 4bab12e8bea70ada9a7024f9166f9109)
Reference:
If you use this data in a publication please cite as:
Andrew Kae*, Kihyuk Sohn*, Honglak Lee, and Erik Learned-Miller.
Augmenting CRFs with Boltzmann Machine Shape Priors for Image Labeling.
Computer Vision and Pattern Recognition, 2013.
*The first and second authors made equal contributions and should be considered co-first authors.
[pdf] [project page]
BibTeX entry:
@inproceedings{GLOC_CVPR13,
author = {Andrew Kae and Kihyuk Sohn and Honglak Lee and Erik Learned-Miller},
title = {Augmenting {CRF}s with {B}oltzmann Machine Shape Priors for Image Labeling},
booktitle = CVPR,
institution = {University of Massachusetts Amherst and University of Michigan Ann Arbor},
year = 2013}
Contact:
Questions and comments can be sent to:
Andrew Kae - akae@cs.umass.edu
Support:
This work was supported by NSF Grant IIS-0916555 and a Google Faculty Research Award. We would like to thank Jerod Weinman and Gary Huang for their source code and helpful discussions. We would also like to thank Ariel Prabawa, Flavio Fiszman, Jonathan Tiao, and Sammy Hajalie for their help in constructing this database.