Object-Centric Spatial Pooling for Image Classification

By Olga Russakovsky, Yuanqing Lin, Kai Yu, Li Fei-Fei. 2012

213 citations recorded by the source corpus.

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

In Media Technology

Citations in the corpus

Extreme Clicking for Efficient Object Annotation (2017, background): “This setting is very cheap in terms of annotation time, but it produces lower quality object detectors, typically performing at only about half the level of accuracy achieved by training from bounding boxes [5, 11, 14, 55, 67].”
HCP: A Flexible CNN Framework for Multi-Label Image Classification (2014, background): “The cross-hypothesis max-pooling in the fusion layer can be formulated as where v ( j ) can be considered as the predicted value for the j th category of the given image.”
Weakly Supervised Object Localization with Multi-Fold Multiple Instance Learning (2015, methodology): “, [7], [38].”
Towards Unified Object Detection and Semantic Segmentation (2014, unknown): “In addition, some ambiguous object hypotheses may be difficult to classify from the information within the window/segment alone, but contextual information, such as local context around each object hypothesis and global image-level context, can help [28,34,11].”
DOCK: Detecting Objects by Transferring Common-Sense Knowledge (2018, methodology): “Context has been used for various vision tasks including object detection [31,9,8,33,4,15], semantic segmentation [28], and object discovery [20,10].”
The Cross-Depiction Problem: Computer Vision Algorithms for Recognising Objects in Artwork and in Photographs (2015, background): “Of the many approaches to visual object classification, the bag-of-words (BoW) family [14, 41, 49] is amongst the most widespread.”
Multi-fold MIL Training for Weakly Supervised Object Localization (2014, unknown): “’12 [23] 30.8 25.0 — 3.6 — 26.0 — — — — — — 21.3 29.9 — — — — — — 15.0 Siva and Xiang’11 [29] 13.4 44.0 3.1 3.1 0.0 31.2 43.9 7.1 0.1 9.3 9.9 1.5 [22] are based on external video data for training.”
Fisher and VLAD with FLAIR (2014, methodology): “In contrast, FLAIR allows for fast encoding for VLAD or Fisher vectors on arbitrary boxes, while leaving the possibility intact to benefit from the recent advances in normalization [18, 14] and the advantages offered by fine spatial pooling [25, 21].”
Large-Scale Weakly Supervised Object Localization via Latent Category Learning (2015, background): “To further suppress the background area and improve the quality of the object regions, researchers model intra-class relations to improve the similarity of the regions of within an object class [7]–[9], [27].”
Efficient Object Detection and Segmentation for Fine-Grained Recognition (2013, methodology): “Our approach is based on identifying regions, specific of the categories of interest, at the time of detection.”
We Don’t Need No Bounding-Boxes: Training Object Class Detectors Using Only Human Verification (2016, methodology): “Following the common protocol for WSOL experiments [8, 9, 12, 42, 60], we exclude trainval images that contain only difficult and truncated instances, ending up with 3550 images.”
Spot On: Action Localization from Pointly-Supervised Proposals (2016, methodology): “We draw inspiration from related work on weakly-supervised object detection, e.g., [6,7,8].”
Weakly Supervised Object Localization with Latent Category Learning (2014, methodology): “, exhaustive search [20, 21], multiple instance learning [13,28,34], inter-intra-class modeling [5,9,23,28,29] and topic model [25,30].”
Weakly Supervised Object Localization Using Things and Stuff Transfer (2017, unknown): “In order to reduce the annotation cost, many previous works learn the detector in a weakly supervised setting [8, 9, 4, 10, 11, 12, 13, 14, 15], i.”
WELDON: Weakly Supervised Learning of Deep Convolutional Neural Networks (2016, methodology): “In computer vision, the dominant approach for WSL is the Multiple Instance Learning (MIL) paradigm [9]: an image is considered as a bag of regions, and the model seeks the max scoring instance in each bag [35, 41, 37, 5, 47, 44, 11].”
Y Qualification Main task Y Return N N Y User qualified ? N Feedback Pass ? Pass ? Annotator training Annotating images (2017, background): “This setting is very cheap in terms of annotation time, but it produces lower quality object detectors, typically performing at only about half the level of accuracy achieved by training from bounding boxes [5, 11, 14, 55, 67].”