Trained unary potentials

If you're using any of the material or code on this page, cite our paper Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials .

Data

Dataset	Database split	Compressed probabilities	Probabilities with logistic regression applied
MSRC	split_msrc.zip	msrc_compressed.zip(46M, 84.05% GA)	N/A
VOC 2010	split_voc2010.zip	voc2010_compressed.zip(771M, 22.21% CA)	voc2010_logit_compressed.zip(3.5G, 27.43% CA)

GA	Global accuracy
CA	Class wise accuracy (VOC)

All the data stores probabilities, normalized per pixel to sum up to 1. If you need access to the raw output of the boosting classifier train them yourself here) or contact the authors. However those raw boosting outputs are a lot larger (MSRC: 3G and VOC: 30G), such that we cannot host them online.

How to use:

Download the source code to load the probability images
Unzip the archive
Load the probabilities as a "ProbImage"

Use "ProbImage::decompress"

See "decompress.cpp" as an example