Author manuscript, published in "International Conference on Computer Vision (2009) 8 p." Combining efficient object localization and image classification

Citations in the corpus

• What Makes for Effective Detection Proposals? (2015, background): “Each of the proposals has its own location bias to diversify the location of the proposals.”
• On the Binding Problem in Artificial Neural Networks (2020, background): “A classifier can then be trained to determine which window contains an object (Rowley et al., 1998; Viola and Jones, 2001; Harzallah et al., 2009).”
• HCP: A Flexible CNN Framework for Multi-Label Image Classification (2014, background): “Indicated by [1], the loss function is defined as J = 1 n n∑ i=1 c∑ j=1 max (0, 1− yijfj(xi))2 (4) where yij = { +1 if xi is annotated with the j th label −1 otherwise.”
• How good are detection proposals, really? (2014, methodology): “In this sense detection proposal methods are related to cascade methods [8, 18, 31], which use a fast (but inaccurate) classifier to discard the vast majority of unpromising proposals.”
• Fisher and VLAD with FLAIR (2014, background): “Sub-window search [15] and selective search [24] opened the door to the use of locality with BoW, which is computationally more expensive than HOG [8] but superior in the quality of the semantic interpretation [15, 27, 12, 24, 25, 5].”
• Segmentation as selective search for object recognition (2011, methodology): “To show that our selective search strategy enables the use of more expensive and potentially more powerful features, we use Bag-of-Words for object recognition [13, 15, 27].”
• Multilabel Image Classification With Regional Latent Semantic Dependencies (2016, methodology): “We also compute the mean average precision (mAP) for the comparisons. op = ∑c i=1N c i∑c i=1N p i , or = ∑c i=1N c i∑c i=1N g i , (15) cp = 1 c c∑ i=1 N ci Npi , cr = 1 c c∑ i=1 N ci Ngi , (16) where c is the number of total labels, N ci is the number of images that are accurately labeled for ith la”
• Selective Search for Object Recognition (2016, methodology): “Much of the related work to object recognition with respect to brute-force search is using the sliding window technique [12].”
• 50 Years of object recognition: Directions forward (2013, background): “Similarly to the work in [360, 361], the authors use a cascade of progressively more costly but more accurate kernels (linear, quasi-linear and non-linear kernels) to efficiently localize the objects.”
• Modeling mutual context of object and human pose in human-object interaction activities (2010, methodology): “In the recent Pascal VOC challenge dataset [9], the difference between context based methods and sliding window based methods for object detection (e.g. detecting bicycles) is only within a small margin of 3− 4% [7, 13].”
• UvA-DARE ( Digital Academic Repository ) The visual extent of an object : suppose we know the object locations (2011, methodology): “Indeed, the work of Harzallah et al. (2009) combined a system for object detection, i.e. localising and classifying objects within images, and a Bag-of-Words object classification system by fusing their respective classifier outputs.”
• Co-localization in Real-World Images (2014, background): “Most state-of-the-art methods require extensive guidance in training, using large numbers of images with human-annotated bounding boxes [11, 28].”
• Assessing the aesthetic quality of photographs using generic image descriptors (2011, methodology): “The proposed image descriptors have been successfully applied to semantic tasks such as object/scene retrieval [25] [21], image classification/annotation [2] [22] and object localization [10].”
• End-to-End Training of Object Class Detectors for Mean Average Precision (2016, methodology): “Earlier approaches [19,22,23,24] apply the classifier to all windows in a dense regular grid, while more recently, object proposal methods [6,25] have been used to greatly reduce the number of windows [7,3].”
• Contextualizing Object Detection and Classification (2015, methodology): “Harzallah et al. [23] introduced the pioneering work for object detection and classification contextualization through the post-processing of probability combination.”
• The Visual Extent of an Object (2011, methodology): “Indeed, the work of Harzallah et al. (2009) combined a system for object detection, i.e. localising and classifying objects within images, and a Bag-of-Words object classification system by fusing their respective classifier outputs.”
• Measuring the Objectness of Image Windows (2011, background): “Many modern class-specific object detectors [12], [18], [25]”
• Reinforcement Learning for Visual Object Detection (2016, background): “Motivated by these limitations, other authors have proposed to use cascades of classifiers[39] [11] [15] [30], to progressively narrow the search space, where weak but fast classifiers are applied early to eliminate image regions unlikely to contain the target, while investing computational…”