Results 1 -
5 of
5
Image Parsing with Graph Grammars and Markov Random Fields Applied to Facade Analysis
"... Existing approaches to parsing images of objects featur-ing complex, non-hierarchical structure rely on exploration of a large search space combining the structure of the object and positions of its parts. The latter task requires random-ized or greedy algorithms that do not produce repeatable resul ..."
Abstract
-
Cited by 2 (0 self)
- Add to MetaCart
(Show Context)
Existing approaches to parsing images of objects featur-ing complex, non-hierarchical structure rely on exploration of a large search space combining the structure of the object and positions of its parts. The latter task requires random-ized or greedy algorithms that do not produce repeatable results or strongly depend on the initial solution. To address the problem we propose to model and optimize the structure of the object and position of its parts separately. We encode the possible object structures in a graph grammar. Then, for a given structure, the positions of the parts are inferred using standard MAP-MRF techniques. This way we limit the application of the less reliable greedy or randomized optimization algorithm to structure inference. We apply our method to parsing images of building facades. The results of our experiments compare favorably to the state of the art. 1.
Team GALEN Organ Modeling through Extraction, Representation and Understanding of Medical Image Content
"... 3.1. Structured coupled low- and high-level visual perception 3 ..."
(Show Context)
Open internship positons in the IMAGINE Group (spring-summer 2013) The IMAGINE Group
"... and constraint programming. In particular, IMAGINE has been working for several years on dense multi-view stereovision. One of the main focuses of the group has been on high precision 3D surface reconstruction from images, targeting large-scale data sets taken under uncontrolled conditions. This exp ..."
Abstract
- Add to MetaCart
(Show Context)
and constraint programming. In particular, IMAGINE has been working for several years on dense multi-view stereovision. One of the main focuses of the group has been on high precision 3D surface reconstruction from images, targeting large-scale data sets taken under uncontrolled conditions. This expertise as well as software have been transferred in 2011 to the startup company Acute3D, powering Autodesk’s 123D Catch (formerly project Photofly), a web service to create 3D models from photographs. Other on-going work include the automatic semantization of 2D and 3D scenes as well as large-scale learning and optimization. Open internship positons (spring-summer 2013) Below is a list of internships open in the IMAGINE Group in spring-summer 2013. Note that only a subset of these positions will be filled, depending on the profile of applicants. No. INT13-01: [MSc] Adaptive windows in stereo correspondence No. INT13-02: [MSc] Comparison of stereo rectification methods No. INT13-03: [MSc] Plausible reconstruction of partially occluded polygonal surfaces in single view 3D range data for indoor scene reconstruction No. INT13-04: [MSc] Appariement probabiliste de bases de données
A MRF Shape Prior for Facade Parsing with Occlusions Supplementary Material
"... In this document we present: • Proof of equivalence of the grid pattern prior to the adjacency pattern prior; • Derivation of the optimization algorithm from section 3; • Confusion matrices corresponding to the results of experiments presented in the paper. 2. An adjacency pattern equivalent to a gi ..."
Abstract
- Add to MetaCart
In this document we present: • Proof of equivalence of the grid pattern prior to the adjacency pattern prior; • Derivation of the optimization algorithm from section 3; • Confusion matrices corresponding to the results of experiments presented in the paper. 2. An adjacency pattern equivalent to a given grid pattern As stated in section 2.1 of the paper, for any grid pattern shape prior, G = (C,R,H,V) there exists an adjacency pattern AG = (SG, V G, HG) encoding the same set of shapes. The set of pixel classes of the adjacency pattern is SG = R × C. We denote the row-class component of a pixel class s = (rs, cs) by r(s) = rs and its column-class component by c(s) = cs. The sets of allowed classes of adjacent pixels are defined as follows (equation 1 of the main paper):
A novel approach to 2D drawings-based reconstruction of 3D building digital models
"... There is an opportunity to significantly impact building energy efficiency through BIM-enabled, simulation-intensive, cost-effective renovation actions. One major hurdle however is the lack of 3D digital models for the majority of existing buildings. Cost-effective and widely applicable methods and ..."
Abstract
- Add to MetaCart
(Show Context)
There is an opportunity to significantly impact building energy efficiency through BIM-enabled, simulation-intensive, cost-effective renovation actions. One major hurdle however is the lack of 3D digital models for the majority of existing buildings. Cost-effective and widely applicable methods and tools are required for the reconstruction of 3D digital models from available information. In this scope, this paper presents ongoing research about the development of tools for semi-automated 3D building model generation from 2D scanned plans. More specifically, the paper focuses on the description of an innovative reconstruction process based on a combination of automated processing and punctual, software-assisted, user intervention. The paper also gives an account of the experiments lead with a research prototype, tested on 90 real architectural floor plans. The results are encouraging and suggest that the mix of software-assistance and focused human intervention may be the best trade-off to upgrade the quality of the generated models and to achieve cost-effectiveness.
