DeepID-Net: multi-stage and deformable deep convolutional neural networks for object detection

By Wanli Ouyang, Ping Luo, Xingyu Zeng, Shi Qiu, Yonglong Tian, Hongsheng Li, Shuo Yang, Zhe Wang, Yuanjun Xiong, Chen Qian and 5 more. 2014

165 citations recorded by the source corpus.

In this paper, we propose multi-stage and deformable deep convolutional neural networks for object detection. This new deep learning object detection diagram has innovations in multiple aspects. In the proposed new deep architecture, a new deformation constrained pooling (defpooling) layer models the deformation of object parts with geometric constraint and penalty. With the proposed multistage training strategy, multiple classifiers are jointly optimized to process samples at different difficulty levels. A ne w pre-training strategy is proposed to learn feature represe ntations more suitable for the object detection task and with good generalization capability. By changing the net structures, training strategies, adding and removing some key components in the detection pipeline, a set of models with large diversity are obtained, which significantly improves the effectiveness of modeling averaging. The proposed approach ranked #2 in ILSVRC 2014. It improves the mean averaged precision obtained by RCNN, which is the stateof-the-art of object detection, from31% to 45%. Detailed component-wise analysis is also provided through extensiv e experimental evaluation.

Subfields: Computer Vision and Pattern Recognition, Industrial and Manufacturing Engineering

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

1. Xiaogang Wang
2. Luc Van Gool
3. Dacheng Tao
4. Yi Yang
5. Ming-Hsuan Yang
6. Trevor Darrell
7. Alan L. Yuille
8. Bernt Schiele

In Industrial and Manufacturing Engineering

1. Xiaogang Wang
2. Wanli Ouyang
3. Trevor Darrell
4. Li Fei-Fei
5. Hongsheng Li
6. Olga Russakovsky
7. Hao Su
8. Zhen Lei

Citations in the corpus

• Object Detection via a Multi-region and Semantic Segmentation-Aware CNN Model (2015, methodology): “On [26], they introduce a deep CNN with a novel deformation constrained pooling layer, a new strategy for pre-training that uses the bounding box annotations provided from ImageNet localization task, and contextual features derived by applying a pre-trained on ImageNet CNN on the whole image and tre”
• ImageNet Large Scale Visual Recognition Challenge (2014, background): “7 The Chinese University of Hong Kong Wanli Ouyang, Ping Luo, Xingyu Zeng, Shi Qiu, Yonglong Tian, Hongsheng Li, Shuo Yang, Zhe Wang, Yuanjun Xiong, Chen Qian, Zhenyao Zhu, Ruohui Wang, Chen-Change Loy, Xiaogang Wang, Xiaoou Tang (Ouyang et al. 2014; Ouyang and Wang 2013) DeepCNet 17.”
• Scalable, High-Quality Object Detection (2014, neutral): “We are deeply sorry if our reported results caused any confusion in the community. entries [20, 15, 19] of the 2014 Imagenet object detection competition [16], use salience-based object localization.”
• Recent advances in convolutional neural networks (2015, background): “Since 2012, the huge success of CNNs in ImageNet challenge [6] rekindles interest in CNN-based object detection [121–127].”
• CRAFT Objects from Images (2016, methodology): “[24, 36] find that using the CNN to shrink the proposals generated by grouping style proposals leads to performance gain.”
• DenseBox: Unifying Landmark Localization with End to End Object Detection (2015, background): “However, most state-of-the-art object detection approaches [26, 20, 8, 14, 41] rely on R-CNN, which divides detection into two steps: salient object proposal generation and region proposal classification.”
• Cross-modal adaptation for RGB-D detection (2016, methodology): “2) Scoring region proposals: This is typically done by using CNNs [15], [22], [37], [46].”
• Object detection by labeling superpixels (2015, result): “This observation is in consistent with the bounding box rejection in [39].”
• Saliency detection by multi-context deep learning (2015, methodology): “Apart from the Clarifai model, there are also other contemporary models such as AlexNet [29], NIN [33], OverFeat [46], DeepID-Net [42], and GoogLeNet [50].”
• Modeling local and global deformations in Deep Learning: Epitomic convolution, Multiple Instance Learning, and sliding window detection (2015, background): “[32], however, recent works on combining convolutional networks [39], or sliding window detectors with CNN features [13, 33, 44] still lag substantially behind the current state-of-the-art techniques [11, 15, 30] that use region proposals delivered, e.”
• SpotGarbage: smartphone app to detect garbage using deep learning (2016, background): “Convolutional Neural Networks (CNNs) have been pivotal in giving state-of-the-art performance for image recognition on the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) [13, 23, 18, 22, 10], scene labeling [8, 20], semantic segmentation [15], and material classification [3].”