Multi-Class Model Fitting by Energy Minimization and Mode-Seeking
Daniel Barath
(MTA SZTAKI Budapest, Hungary)
Abstract:
We propose a novel method, called Multi-X, for general
multi-class multi-instance model fitting – the problem of
interpreting input data as a mixture of noisy observations
originating from multiple instances of multiple types. The
proposed approach combines global energy optimization
and mode-seeking in the parameter domain. Considering
an outlier class makes it robust. Key optimization parameters
like the outlier threshold are set adaptively within the
algorithm. A cross-validation-based step removes unstable,
and thus insignificant, instances.
Multi-X outperforms significantly the state-of-the-art on
the standard AdelaideRMF (mutiple plane segmentation,
multiple rigid motion detection) and Hopkins datasets (motion
segmentation) and in experiments on 3D LIDAR data
(simultaneous plane and cylinder fitting) and on 2D edge
interpretation (circle and line fitting). Multi-X runs in time
approximately linear in the number of data points at around
0.1 second per 100 points, an order of magnitude faster than
available implementations of commonly used methods.