Contextual classification with functional Max-Margin Markov Networks

By Daniel Munoz, Andrew Bagnell. 2009

363 citations recorded by the source corpus.

We address the problem of label assignment in computer vision: given a novel 3D or 2D scene, we wish to assign a unique label to every site (voxel, pixel, superpixel, etc.). To this end, the Markov Random Field framework has proven to be a model of choice as it uses contextual information to yield improved classification results over locally independent classifiers. In this work we adapt a functional gradient approach for learning high-dimensional parameters of random fields in order to perform discrete, multi-label classification. With this approach we can learn robust models involving high-order interactions better than the previously used learning method. We validate the approach in the context of point cloud classification and improve the state of the art. In addition, we successfully demonstrate the generality of the approach on the challenging vision problem of recovering 3-D geometric surfaces from images.

Subfields: Computer Vision and Pattern Recognition, Environmental Engineering

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

In Environmental Engineering

Citations in the corpus

DIMENSIONALITY BASED SCALE SELECTION IN 3D LIDAR POINT CLOUDS (2012, background): “MMS : Datasets over two urban areas (France and United States, respectively MMSF and MMSU ) from distinct mobile mapping systems have been processed (Munoz et al., 2009).”
Learning message-passing inference machines for structured prediction (2011, methodology): “We observe that the best approach overall is the functional gradient M(3)N approach of [14].”
The Hessigheim 3D (H3D) Benchmark on Semantic Segmentation of High-Resolution 3D Point Clouds and Textured Meshes from UAV LiDAR and Multi-View-Stereo (2021, background): “Although Mobile Laser Scanning (MLS) point clouds of typical urban scenes are often provided as stand-alone products (Roynard et al., 2018; Munoz et al., 2009; Hackel et al., 2017), the concurrent availability of LiDAR data and imagery in the form of meshes is often pursued (Riemenschneider et al.,…”
Incremental dense semantic stereo fusion for large-scale semantic scene reconstruction (2015, background): “incrementally [12] or rely on LIDAR data [24].”
Collective Stability in Structured Prediction: Generalization from One Example (2013, background): “…(Taskar et al., 2002; Richardson & Domingos, 2006; Sen et al., 2008); in computer vision, object classifiers are often trained on a handful of large outdoor scenes (Munoz et al., 2009); in cross-document coreference resolution, the training data may be a single, large corpus (Singh et al., 2010).”
Performance of histogram descriptors for the classification of 3D laser range data in urban environments (2012, methodology): “Both local classifiers are compared with a collective classification approach using Conditional Random Fields (CRF)— the Functional Max-Margin Markov Networks (FM(3)N) [4].”
Deep Learning for LiDAR Point Clouds in Autonomous Driving: A Review (2020, neutral): “Oakland 3-D Point Cloud Dataset [57].”
Learning to search: Functional gradient techniques for imitation learning (2009, methodology): “That paper, and later (Munoz et al. 2008; Munoz et al. 2009 ), demonstrate that the subgradient method and its functional gradient generalizations perform well on large scale LADAR classification problems that require approximate inference.”
SUM: A Benchmark Dataset of Semantic Urban Meshes (2021, background): “The Oakland 3D [16] is one of the earliest mobile laser scanning (MLS) point cloud datasets, which was designed for classification of outdoor scenes.”
The H3D Dataset for Full-Surround 3D Multi-Object Detection and Tracking in Crowded Urban Scenes (2019, background): “Net [13] and Oakland dataset [16] are two point cloud datasets that provide semantic labels for point cloud classification.”
Sensor fusion for semantic segmentation of urban scenes (2015, methodology): “For point clouds, the same bottom-up pipeline is typically taken [10] [11] [12] [13] [14], starting with an oversegmentation, followed by regional feature extraction, classification, and a CRF.”
Harmony Potentials (2011, background): “Another way of exploiting global image information is by inferring 3D scene geometry to discover where objects are likely to appear and how big they can be (Hoiem et al. 2007; Hoiem et al. 2008; Munoz et al. 2009).”
Learning to search: structured prediction techniques for imitation learning (2009, background): “…and functional gradient approaches to structured learning demonstrate stateof-the-art performance on a diverse set of problems ranging from optical character recognition to LADAR point cloud classification (Ratliff, Bagnell, & Zinkevich, 2007a; Munoz, Vandapel, & Hebert, 2008; Munoz et al., 2009).”
Large-Scale Point Cloud Semantic Segmentation with Superpoint Graphs (2017, methodology): “The solution is then spatially regularized using graphical models [37, 24, 34, 44, 21, 2, 38, 35, 49] or structured optimization [27].”
Context tracker: Exploring supporters and distracters in unconstrained environments (2011, neutral): “Context information has been applied actively in object detection [9], object classification [20, 23], object recognition [25].”
A search-classify approach for cluttered indoor scene understanding (2012, methodology): “Markov Random Fields were used in this context by computing the classification of a point from its local descriptor and neighbors [Anguelov et al. 2005; Munoz et al. 2009].”
Paris-Lille-3D: A large and high-quality ground-truth urban point cloud dataset for automatic segmentation and classification (2017, methodology): “The Oakland 3-D Point Cloud Dataset11 (Munoz et al., 2009) dataset was acquired by a MLS system mounted with a side-looking Sick monofiber LiDAR.”