DeepID-Net: Deformable deep convolutional neural networks for object detection

By Wanli Ouyang, Xiaogang Wang, Xingyu Zeng, Shi Qiu, Ping Luo, Yonglong Tian, Hongsheng Li, Shuo Yang, Zhe Wang, Chen Change Loy and 1 more. Computer Vision and Pattern Recognition, 2014

450 citations recorded by the source corpus.

In this paper, we propose deformable deep convolutional neural networks for generic object detection. This new deep learning object detection framework has innovations in multiple aspects. In the proposed new deep architecture, a new deformation constrained pooling (def-pooling) layer models the deformation of object parts with geometric constraint and penalty. A new pre-training strategy is proposed to learn feature representations 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 model averaging. The proposed approach improves the mean averaged precision obtained by RCNN [14], which was the state-of-the-art, from 31% to 50.3% on the ILSVRC2014 detection test set. It also outperforms the winner of ILSVRC2014, GoogLeNet, by 6.1%. Detailed component-wise analysis is also provided through extensive experimental evaluation, which provide a global view for people to understand the deep learning object detection pipeline.

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

View Paper: https://www.semanticscholar.org/paper/c705e85be5922d7ffa60418d49e75aeedf065a75

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 Industrial and Manufacturing Engineering

Citations in the corpus

Deep Learning for Generic Object Detection: A Survey (2018, methodology): “The second is based on deformable part-based model [72], which finds the max response to a part filter with spatial constraint taken into consideration [199, 84, 270].”
Quantization Mimic: Towards Very Tiny CNN for Object Detection (2018, background): “The target of object detection [5,6,7,8,9,10] is to locate and classify the objects in images.”
Factors in Finetuning Deep Model for Object Detection with Long-Tail Distribution (2016, methodology): “By default, we use the bounding box rejection approach in [25] so that around 6% candidate regions from selective search are retained.”
Saliency detection by multi-context deep learning (2015, methodology): “Notably, best performing entries [29, 46, 59, 50, 42, 43] on ImageNet ILSVRC [14] and PASCAL VOC [15] benchmarks are all variants of deep CNNs since 2012.”
Accelerating magnetic resonance imaging via deep learning (2016, background): “Especially, convolutional neural network (CNN) has exhibited its significance in addressing large-scale vision tasks such as object detection [8] and super-resolution [9].”
Automatic segmentation of nine retinal layer boundaries in OCT images of non-exudative AMD patients using deep learning and graph search. (2017, background): “In recent years, deep learning [34–36] based neural networks have been demonstrated to be a very powerful tool in the field of computer vision [37,38].”
Deep Feature Flow for Video Recognition (2016, methodology): “8% is reported in [22], by combining CRAFT [46] and DeepID-net [30] object detectors trained on the ImageNet data, using both VGG-16 [39] and GoogleNet-v2 [20] models, with various tricks (multi-scale training/testing, adding context information, model ensemble).”
Deep Learning Strong Parts for Pedestrian Detection (2015, methodology): “taking the full images as input, such as [14] (strategy 2), and (3) pre-training the deep models by using ImageNet training data with object-level annotations of 1000 classes [25], i.”
Object Detection in Videos with Tubelet Proposal Networks (2017, background): “…GoogLeNet [27], VGG [25] and ResNet [8], were proposed to improve the learning capability on large-scale computer vision datasets for various computer vision tasks, such as object detection [5, 22, 21, 19], semantic segmentation [18, 2, 15], tracking [29, 1, 30], scene understanding [23, 24] etc .”
Discover and Learn New Objects from Documentaries (2017, background): “Recent years have witnessed a wave of innovation in object detection driven by the advances in deep learning [12, 29, 35].”
Chained Cascade Network for Object Detection (2017, background): “Deep models were found to be effective in object detection [11, 10, 12, 24, 25, 18, 21, 22, 41, 19, 19, 20].”
Learning fine-grained features via a CNN Tree for Large-scale Classification (2015, background): “Convolutional neural networks (CNN) [1], [2] have recently demonstrated superior performance on many tasks such as image classification [3], [4], [5], object detection [6], [7], [8], [9], [10], [11], object tracking [12], [13], [14], text detection [15], [16], text recognition [17], [18], [19], vide”
Learning Affective Features With a Hybrid Deep Model for Audio–Visual Emotion Recognition (2018, background): “Due to the large-scale available training data and the effective training strategies introduced in recent works, CNNs have exhibited signiﬁcant success in various vision tasks like object detection and recognition [29]–[32].”
Understand scene categories by objects: A semantic regularized scene classifier using Convolutional Neural Networks (2015, methodology): “In DeepIDNet [15], a refinement scheme is conducted to refine the object detection using the image classification result.”
Learning Chained Deep Features and Classifiers for Cascade in Object Detection (2017, result): “We compare our framework with several other state-of-art approaches [13, 28, 15, 18, 32, 14, 35].”
Exploit All the Layers: Fast and Accurate CNN Object Detector with Scale Dependent Pooling and Cascaded Rejection Classifiers (2016, unknown): “With the exceptional power on image classification, CNN has been applied to object detection and achieves promising results [34, 14, 7, 44, 13, 27].”