On Taxonomies for Multi-class Image Categorization
By Alexander Binder, K. Müller, Motoaki Kawanabe. 2012
57 citations recorded by the source corpus.
We study the problem of classifying images into a given, pre-determined taxonomy. This task can be elegantly translated into the structured learning framework. However, despite its power, structured learning has known limits in scalability due to its high memory requirements and slow training process. We propose an efficient approximation of the structured learning approach by an ensemble of local support vector machines (SVMs) that can be trained efficiently with standard techniques. A first theoretical discussion and experiments on toy-data allow to shed light onto why taxonomy-based classification can outperform taxonomy-free approaches and why an appropriately combined ensemble of local SVMs might be of high practical use. Further empirical results on subsets of Caltech256 and VOC2006 data indeed show that our local SVM formulation can effectively exploit the taxonomy structure and thus outperforms standard multi-class classification algorithms while it achieves on par results with taxonomy-based structured algorithms at a significantly decreased computing time.
Subfields: Computer Vision and Pattern Recognition, Media Technology
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
- K. Müller
- Wojciech Samek
- Grgoire Montavon
- Luc Van Gool
- Andrea Vedaldi
- Alexander Binder
- Dacheng Tao
- Trevor Darrell
In Media Technology
- Matthew B. Blaschko
- Arthur Gretton
- Wojciech Zaremba
- K. Müller
- Wojciech Samek
- Motoaki Kawanabe
- Marius Kloft
- Alexander Binder
Citations in the corpus
- Sparselet Models for Efficient Multiclass Object Detection (2012, unknown): “Others optimize a discriminative objective [9, 10].”
- Enhanced representation and multi-task learning for image annotation (2013, background): “Future work will study a holistic learning procedure where representations and transfer between tasks are part of an overall integrated optimization as well as the impact of both ideas to more complex semantic ranking measures like the ones in [42].”
- Taxonomic Prediction with Tree-Structured Covariances (2013, methodology): “[2] made use of a learning framework that is perhaps the most similar to that employed here, based on structured output prediction.”
- Insights from Classifying Visual Concepts with Multiple Kernel Learning (2011, background): “knowledge from taxonomies for multi-label ranking [60,61] or the classical multi-class classification, semantic information or spatial priors.”