Towards Rich Feature Discovery With Class Activation Maps Augmentation for Person Re-Identification

Yang, Wenjie; Huang, Houjing; Zhang, Zhang; Chen, Xiaotang; Huang, Kaiqi; Shu, Zhang

doi:10.1109/cvpr.2019.00148

Cited by 232 publications

(95 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Nevertheless, none of the existing ReID models addresses omni-scale feature learning. In recent years, deep convolutional neural networks (CNNs) have been widely used in person ReID to learn discriminative features [2,27,31,42,47,63,65,77]. However, most of the CNNs adopted, such as ResNet [13], were originally designed for object category-level recognition tasks that are fundamentally different from the instance-level recognition task in ReID.…”

Section: Introductionmentioning

confidence: 99%

Omni-Scale Feature Learning for Person Re-Identification

Zhou

Yang

Cavallaro

et al. 2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

664

351

View full text Add to dashboard Cite

As an instance-level recognition problem, person reidentification (ReID) relies on discriminative features, which not only capture different spatial scales but also encapsulate an arbitrary combination of multiple scales. We callse features of both homogeneous and heterogeneous scales omni-scale features. In this paper, a novel deep ReID CNN is designed, termed Omni-Scale Network (OSNet), for omni-scale feature learning. This is achieved by designing a residual block composed of multiple convolutional feature streams, each detecting features at a certain scale. Importantly, a novel unified aggregation gate is introduced to dynamically fuse multiscale features with input-dependent channel-wise weights. To efficiently learn spatial-channel correlations and avoid overfitting, the building block uses both pointwise and depthwise convolutions. By stacking such blocks layerby-layer, our OSNet is extremely lightweight and can be trained from scratch on existing ReID benchmarks. Despite its small model size, our OSNet achieves state-ofthe-art performance on six person-ReID datasets. Code and models are available at: https://github.com/ KaiyangZhou/deep-person-reid.

show abstract

Section: Introductionmentioning

confidence: 99%

Omni-Scale Feature Learning for Person Re-Identification

Zhou

Yang

Cavallaro

et al. 2019

2019 IEEE/CVF International Conference on Computer Vision (ICCV)

664

351

View full text Add to dashboard Cite

show abstract

“…We compare the proposed method with 33 recent published works including (1) global feature based methods which aims to learn the global feature from the feature map directly, including PAN [74], DMML [7], DCDS [1], VCFL [30], MVPM [41], LRDNN [79], RB [35], LITM [63], IANet [23], Sphere [14], BNNeck [32], OSNet [78], AANet [46], DG-Net [72], BDB [12], Circle [42], SFT [31], (2) part based methods including PCB+RPP [43], Local [57], HPM [16], CASN [71], AutoReID [34], MGN [49], BHP [20] and Pyramidal [68] which utilize the semantic parts or horizontal stripes to extract part-level feature, and (3) attention based methods including MHAN [3], CAMA [58], SONA [53], CAR [80], SCAL [6], ABD-Net [8], DAAF [10] and RGA [65]. These methods are categorized into 3 types based on different backbones: the ones which employ ResNet-50 directly, the ones which modify ResNet-50 by introducing additional branches, attention subnets or dilated convolution, and the others which don't use ResNet-50.…”

Section: Comparison Resultsmentioning

confidence: 99%

“…Specifically, they aim to learn the features by improving loss functions [9,14,22,31,41,42,50,55,63], improving the training techniques [1,4,12,24,32,35,37,54], adding additional network modules [23,23,51,62], using extra semantic annotations [30,46,47,79] or generating more training samples [17,33,72,76,77]. Besides, more recent studies [3,6,8,10,27,28,38,46,48,53,58,61,64,65,67,80] integrate attention mechanisms into deep models to enhance the feature representation. To obtain the holistic features, most of these methods utilize global average pooling (GAP), global max pooling (GMP) or both of them on each channel of the feature map.…”

Section: Related Workmentioning

confidence: 99%

Discriminative Spatial Feature Learning for Person Re-Identification

Peng

Huang

Wang

et al. 2020

Proceedings of the 28th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Person re-identification (ReID) aims to match detected pedestrian images from multiple non-overlapping cameras. Most existing methods employ a backbone CNN to extract a vectorized feature representation by performing some global pooling operations (such as global average pooling and global max pooling) on the 3D feature map (i.e., the output of the backbone CNN). Although simple and effective in some situations, the global pooling operation only focuses on the statistical properties and ignores the spatial distribution of the feature map. Hence, it can not distinguish two feature maps when they have similar response values located in totally different positions. To handle this challenge, a novel method is proposed to learn the discriminative spatial features. Firstly, a self-constrained spatial transformer network (SC-STN) is introduced to handle the misalignments caused by detection errors. Then, based on the prior knowledge that the spatial structure of a pedestrian often keeps robust in vertical orientation of images, a novel vertical convolution network (VCN) is proposed to extract the spatial feature in vertical. Extensive experimental evaluations on several benchmarks demonstrate that the proposed method achieves state-of-theart performances by introducing only a few parameters to the backbone.

show abstract

“…Because of the modularity of our work, this model could replace the extraction and classification steps without changing the acquisition and presentation results. Another direction for expansion is the usage of architectural changes, such as attention operations (Prabhakararao and Dandapat 2020), or post hoc interpretation tools, such as class activation maps (Yang et al 2019), which allow the visualisation of the features that have led the network to produce a specific diagnostic. Although current technologies are increasingly black-box given the advent of DL, the application of these post-hoc analysis tools can have a multiplicative effect in a man-plus-machine scenario, especially through the interactivity of AR, which would allow the visualisation of the decision maps and, thus, identify whether the classifier is confidently producing a prediction (Strodthoff et al 2020).…”

Section: Discussionmentioning

confidence: 99%

Towards assisted electrocardiogram interpretation using an AI-enabled Augmented Reality headset

Lampreave

Jiménez-Pérez

Sanz

et al. 2020

Computer Methods in Biomechanics and Biomedical Engineering: Im

View full text Add to dashboard Cite

The interpretation of electrocardiograms (ECGs) is key for the diagnosis and monitoring of cardiovascular health. Despite the progressive digital transformation in healthcare, it is still common for clinicians to analyse ECG printed on paper. Although some systems provide signal processing-based ECG classification, clinicians often find it unreliable. Artificial Intelligence (AI) techniques are becoming state-of-the-art for ECG processing but the lack of digitised ECG has hampered the clinical translation of these techniques. Concurrently, we are living a rise in augmented reality (AR) technologies, with an increasing availability of devices. In this work, we present an automatic digitisation and assisted interpretation of ECG based on an AI-enabled Augmented Reality headset. The AR headset is used to acquire an image of the printed ECG, from which the digitised ECG signal is extracted. Afterwards, the digitised ECG is introduced into a Deep Learning (DL) algorithm pre-trained on a public database of 12-lead ECG recordings. The output of the DL algorithm classifies the ECG signal onto different cardiomyopathy categories, which is then visualized back in the AR headset. Preliminary classification results on simulated ECG images (96.5% of accuracy) confirm the potential of the developed approach to contribute on the digital transformation of ECG processing.

show abstract

Towards Rich Feature Discovery With Class Activation Maps Augmentation for Person Re-Identification

Cited by 232 publications

References 32 publications

Omni-Scale Feature Learning for Person Re-Identification

Omni-Scale Feature Learning for Person Re-Identification

Discriminative Spatial Feature Learning for Person Re-Identification

Towards assisted electrocardiogram interpretation using an AI-enabled Augmented Reality headset

Contact Info

Product

Resources

About