HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection

By Tao Kong, Anbang Yao, Yurong Chen, Fuchun Sun. 2016

853 citations recorded by the source corpus.

Almost all of the current top-performing object detection networks employ region proposals to guide the search for object instances. State-of-the-art region proposal methods usually need several thousand proposals to get high recall, thus hurting the detection efficiency. Although the latest Region Proposal Network method gets promising detection accuracy with several hundred proposals, it still struggles in small-size object detection and precise localization (e.g., large IoU thresholds), mainly due to the coarseness of its feature maps. In this paper, we present a deep hierarchical network, namely HyperNet, for handling region proposal generation and object detection jointly. Our HyperNet is primarily based on an elaborately designed Hyper Feature which aggregates hierarchical feature maps first and then compresses them into a uniform space. The Hyper Features well incorporate deep but highly semantic, intermediate but really complementary, and shallow but naturally high-resolution features of the image, thus enabling us to construct HyperNet by sharing them both in generating proposals and detecting objects via an end-to-end joint training strategy. For the deep VGG16 model, our method achieves completely leading recall and state-of-the-art object detection accuracy on PASCAL VOC 2007 and 2012 using only 100 proposals per image. It runs with a speed of 5 fps (including all steps) on a GPU, thus having the potential for real-time processing.

Subfields: 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 Computer Vision and Pattern Recognition

Citations in the corpus

Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (2019, methodology): “After that, lots of algorithms are proposed to improve its performance, including architecture redesign and reform [4, 9, 5, 28, 30], context and attention mechanism [2, 51, 38, 7, 44], multiscale training and testing [54, 41], training strategy and loss function [40, 52, 61, 17], feature fusion and”
Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks (2015, neutral): “0 HyperNet [21] RPN 07++12 71.”
Single-Shot Refinement Neural Network for Object Detection (2017, methodology): “9 HyperNet[24] VGG-16 ∼ 1000× 600 100 0.”
Scale-Aware Trident Networks for Object Detection (2019, methodology): “Methods like HyperNet [22] and ION [2] concatenate low-level and high-level features from different layers to generate better feature maps for prediction.”
Deep Feature Pyramid Reconfiguration for Object Detection (2018, methodology): “For instance, the HyperNet [27] and ION [3] combine features from multiple layers before making detection.”
Multi-level Fusion Based 3D Object Detection from Monocular Images (2018, methodology): “A region proposal network (RPN) is utilized to generate 2D proposals in the image, as RPN provides strong objectness con-ﬁdence regions of interest (RoIs) with CNN features and it can share weights with the down-stream detection network [31, 36, 37, 19].”
DeepCrack: A deep hierarchical feature learning architecture for crack segmentation (2019, neutral): “Multi-scale deep architectures were also developed, which came into two main directions: (1) inputting images at several scales into the networks [11,29] ; (2) combining feature maps from different layers of a deep architecture [32–34] .”
Cascaded Pyramid Network for Multi-person Pose Estimation (2017, methodology): “In order to improve the efﬁciency and keep integrity of information transmission, our ReﬁneNet transmits the information across different levels and ﬁnally integrates the informations of different levels via upsam-pling and concatenating as HyperNet [23].”
Squeeze and Excitation Rank Faster R-CNN for Ship Detection in SAR Images (2019, methodology): “Then, these layers are concatenated along the channel axis as [13] and [20].”
Recent Advances in Deep Learning for Object Detection (2019, methodology): “9 HyperNet [50] VGG-16 2016 ∼ 600 × 1000 76.”
FoveaBox: Beyond Anchor-based Object Detector (2019, background): “Several attempts have been performed to boost the performance of the detector, including feature pyramid [27][24][21], multiscale [44][34], and object relations [17], etc. Compared to two-stage approaches, the one-stage pipeline skips object proposal generation and predicts bounding boxes and class…”
Towards Large-Scale Small Object Detection: Survey and Benchmarks (2022, background): “HyperNet [66] supposes that the information of a Region of Interest (RoI) could distribute over all layers of backbone network and needs to be well organized.”