Generic Object Detection with Dense Neural Patterns and Regionlets

By Will Y. Zou, Xiaoyu Wang, Miao Sun, Yuanqing Lin. 2014

67 citations recorded by the source corpus.

This paper addresses the challenge of establishing a bridge between deep convolutional neural networks and conventional object detection frameworks for accurate and efficient generic object detection. We introduce Dense Neural Patterns , short for DNPs, which are dense local features derived from discriminatively trained deep convolutional neural networks. DNPs can be easily plugged into conventional detection frameworks in the same way as other dense local features(like HOG or LBP). The effectiveness of the proposed approach is demonstrated with the Regionlets object detection framework. It achieved 46.1% mean average precision on the PASCAL VOC 2007 dataset, and 44.1% on the PASCAL VOC 2010 dataset, which dramatically improves the original Regionlets approach without DNPs. It is the first approach efficiently applying dee p convolutional features for conventional object detection models.

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

Object Detection Networks on Convolutional Feature Maps (2015, methodology): “The mAP on PASCAL VOC 2007 is 45-47% [8], [9], [10], [11] for hybrid methods, and is 47% for R-CNN that just uses SVM on the last convolutional layer.”
Part Detector Discovery in Deep Convolutional Neural Networks (2014, neutral): “Zou et al. [19] solves this by associating the output of a hidden layer with spatially related parts of the input.”
DeepID-Net: multi-stage and deformable deep convolutional neural networks for object detection (2014, background): “First, features should capture the most discriminative information of object classes.”
DeepID-Net: Object Detection with Deformable Part Based Convolutional Neural Networks (2017, unknown): “Because of the power on learning feature representation, deep models have been widely used for object recognition, detection and other vision tasks [1], [6], [7], [19], [20], [21], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47],”
Active Object Localization with Deep Reinforcement Learning (2015, methodology): “[38] adapts the regionlets framework with CNN features too.”
Adaptive region pooling for object detection (2015, background): “However, large appearance variations of training examples lead to less desirable clustering results, where exemplars that are less common can be easily absorbed by the dominant clusters.”
Do More Dropouts in Pool5 Feature Maps for Better Object Detection (2014, methodology): “We evaluate performance on the datasets: Pascal Visual Object Challenge (VOC) 2007 [7] and VOC 2010-2012 [8] [15]; SPP (without combination) [18]; DNP+R [37]; Regionlets [33]; Szegedy et al. [31]; DPM v5 [9]).”
DeepID-Net: Deformable deep convolutional neural networks for object detection (2014, background): “Because of the power on learning feature representation, deep models have been widely used for object recognition, detection and other vision tasks [36, 54, 19, 37, 61, 17, 22, 14, 28, 30, 55, 56, 27, 23, 41, 43, 41, 42, 25, 26, 24, 58].”
Convolutional Channel Features (2015, background): “There have already been some works attempting to connect CNN with popular shallow models, like DPM [22, 44] and Regionlet [57].”
The treasure beneath convolutional layers: Cross-convolutional-layer pooling for image classification (2014, methodology): “This is as opposed to the approach in [12] (although developed for a different application) which treats the fea ture vector from one spatial unit as the local feature.”
Fully Convolutional Networks for Dense Semantic Labelling of High-Resolution Aerial Imagery (2016, methodology): “input image rather than just a patch; the second layer convolutions are applied to the outputs of the rst, and so on resulting in a 2D output matrix of class labels rather than a single classication [32, 28]. The convolutional approach was taken one step further in [17] where it was applied also to ”