Sharing Visual Features for Multiclass and Multiview Object Detection

By Antonio Torralba, Kevin P. Murphy, William T. Freeman. 2007

798 citations recorded by the source corpus.

We consider the problem of detecting a large number of different classes of objects in cluttered scenes. Traditional approaches require applying a battery of different classifiers to the image, at multiple locations and scales. This can be slow and can require a lot of training data since each classifier requires the computation of many different image features. In particular, for independently trained detectors, the (runtime) computational complexity and the (training-time) sample complexity scale linearly with the number of classes to be detected. We present a multitask learning procedure, based on boosted decision stumps, that reduces the computational and sample complexity by finding common features that can be shared across the classes (and/or views). The detectors for each class are trained jointly, rather than independently. For a given performance level, the total number of features required and, therefore, the runtime cost of the classifier, is observed to scale approximately logarithmically with the number of classes. The features selected by joint training are generic edge-like features, whereas the features chosen by training each class separately tend to be more object-specific. The generic features generalize better and considerably reduce the computational cost of multiclass object detection

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

What Makes for Effective Detection Proposals? (2015, neutral): “The authors report top results on ImageNet.”
Cross-Stitch Networks for Multi-task Learning (2016, background): “and their widespread applications in other fields, such as genomics [38], natural language processing [7, 8, 35] and computer vision [3, 10, 30, 31, 40, 51, 53, 58].”
Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials (2011, methodology): “First, the boosted unary classifiers are trained using the JointBoost algorithm [21], using the features described in Section 5.”
Learning to Reconstruct Shapes from Unseen Classes (2018, background): “A number of earlier papers have explored sharing features across categories to recognize new objects from a few examples [Bart and Ullman, 2005, Farhadi et al., 2009, Lampert et al., 2009, Torralba et al., 2007].”
How good are detection proposals, really? (2014, methodology): “Although traditionally used for class specific detection, cascade methods also apply to sets of categories [26], and in principle can be applied to a very large set of categories (see Bing method below).”
Flexible Clustered Multi-Task Learning by Learning Representative Tasks (2016, background): “The new approach allows an arbitrary task to be described by multiple representative tasks, effectively soft-assigning a task to multiple clusters with different weights.”
Avoiding Confusing Features in Place Recognition (2010, methodology): “Finally, the task of confuser detection has some similarities with the task of feature selection in category-level recognition [25,26,27] and retrieval [28,29,30].”
Human Uncertainty Makes Classification More Robust (2019, background): “, sharing of representations [47, 9, 22], effective combination of models [23], or improved accuracy of classification through hierarchical prediction [32, 8].”
Hypergraph spectral learning for multi-label classification (2008, background): “Various approaches have been proposed in the past to exploit the correlation information contained in multiple labels [15, 16, 20, 21, 22, 26, 27].”
Analyzing 3D Objects in Cluttered Images (2012, background): “One can augment viewbased models to share local parts across views [27] [21] [32].”
Zero-shot Learning with Semantic Output Codes (2009, methodology): “In computer vision, techniques for sharing features across object classes have been investigated (Torralba & Murphy, 2007; Bart & Ullman, 2005) but relatively little work has focused on recognizing entirely novel classes, with the exception of Lampert et al. (2009) predicting visual properties of…”
Learning Visual Representations using Images with Captions (2007, background): “In [21] Torralba introduced a discriminative (boosted) learning framework that learns common structure.”
A Unified Framework for Multioriented Text Detection and Recognition (2014, background): “Feature sharing, which leverages common features shared across different object classes to train classifiers jointly, has proven to be very effective in multi-class object detection [23], [24] and recognition [25].”
Distribution Augmentation for Generative Modeling (2020, background): “Training on multiple tasks in NLP (Collobert & Weston, 2008), computer vision (Torralba et al., 2007), and across multiple modalities or datasets (Ngiam et al., 2011) is known to provide helpful inductive biases and improve generalization.”
A Hough transform-based voting framework for action recognition (2010, positive): “Instead of learning a separate classifier for each action, we train a single classifier and benefit from the sharing of features between classes as demonstrated in [33].”