Scale-invariant batch-adaptive residual learning for person re-identification

Sikdar, Arindam; Chowdhury, Ananda S.

doi:10.1016/j.patrec.2019.11.032

Cited by 11 publications

(3 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is possible with the help of supervised distance learning and the Siamese convolution network architecture [56]. Due to its simplicity, easy design and low complexity, this model gains attention among the scientific community and has been successfully adopted for weakly supervised metric learning [38], signature verification [57], person re-identification [58] and face verification [34], where the number of labelled instances per classes in a datasets is not enough to train a traditional CNN classifier. Due to the availability of few training samples and the existence of both pixel-wise and channel-wise correlations, the network is enforced to relearn redundant information during the back-propagation stage, hampering the effective learning of the deep model.…”

Section: Proposed Sicodef 2 Net Modelmentioning

confidence: 99%

SiCoDeF² Net: Siamese Convolution Deconvolution Feature Fusion Network for One-Shot Classification

Roy¹,

Kar²,

Paoletti

et al. 2021

IEEE Access

View full text Add to dashboard Cite

Nowadays, deep convolutional neural networks (CNNs) for face recognition exhibit a performance comparable to human ability in the presence of the appropriate amount of labelled training data. However, training CNNs remains as an arduous task due to the lack of training samples. To overcome this drawback, applications demand one-shot learning to improve the obtained performances over traditional machine learning approaches by learning representative information about data categories from few training samples. In this context, Siamese convolutional network (SiConvNet) provides an interesting deep architecture to tackle the data limitation. In this regard, applying the convolution operation on real world images by using the trainable correlative Gaussian kernel adds correlations to the output images, which hinder the recognition process due to the blurring effects introduced by the convolution kernel application. As a result the pixel-wise and channel-wise correlations or redundancies could appear in both single and multiple feature maps obtained by a hidden layer. In this sense, convolution-based models fail to generalize the feature representation because of both the strong correlations presence in neighboring pixels and the channel-wise high redundancies between different channels of the feature maps, which hamper the effective training. Deconvolution operation helps to overcome the shortcomings that limit the conventional SiConvNet performance, learning successfully correlation-free features representation. In this paper, a simple but efficient Siamese convolution deconvolution feature fusion network (SiCoDeF 2 Net) is proposed to learn the invariant and discriminative complementary features generated from both the (i) subconvolution (SCoNet) and (ii) sub deconvolutional (SDeNet) networks using a concatenation operation which significantly improves the one-shot unconstrained facial recognition task. Extensive experiments performed on several widely used benchmarks, provide promising results, where the proposed SiCoDeF 2 Net model significantly outperforms the current state-of-art in terms of classification accuracy, F1, precision and recall. The code will be available on: https://github.com/purbayankar/SiCoDeF2Net.

show abstract

Section: Proposed Sicodef 2 Net Modelmentioning

confidence: 99%

SiCoDeF² Net: Siamese Convolution Deconvolution Feature Fusion Network for One-Shot Classification

Roy¹,

Kar²,

Paoletti

et al. 2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…At [30,54] it is stated that the residual networks are easy to optimize. However residual networks have less parameter than other deep learning methods [53].…”

Section: Model-4: Residual Networkmentioning

confidence: 99%

Deep Learning Applications for Hyperspectral Imaging: A Systematic Review

Özdemir¹,

Polat²

2020

JIEC

View full text Add to dashboard Cite

Since the acquisition of digital images, scientific studies on these images have been making significant progress. The sizes and quality of the images obtained have increased greatly from past to present. However, when the information contained in these images remains on the visible band (RGB band), the results that can be obtained are limited. For this reason, the need to acquire images with more broadband information has emerged. Hyperspectral Imaging (HSI) method has been developed to meet this need. A hyperspectral image consists of reflections in hundreds of different bands of the electromagnetic spectrum. Each object exhibits a unique reflection characteristic. Due to this characteristic, objects can be separated from each other using hyperspectral imaging. Hyperspectral cameras are used to obtain this image. The information it contains is much more than an RGB image, so deeper results can be achieved than the human eye can see. In this respect, it has great importance. Artificial intelligence technologies are used extensively in image processing as well as in many other fields. As a result, classification studies are carried out on hyperspectral images with machine learning methods. Machine learning methods can be considered as the most general terms of supervised machine learning, unsupervised machine learning, and reinforced machine learning. Supervised machine learning methods mainly: Support Vector Machines(SVM), k-Nearest Neighborhood(k-NN), Decision Trees, Random Forest, Linear Regression and Neural Networks(NN). Neural networks are also used as an unsupervised learning method. However, Deep Learning, a specialized method of artificial neural networks, is highly preferred due to its unique structure. Artificial intelligence methods have been widely used in recent years, especially for the classification of hyperspectral images containing complex information. Considering the studies, it is seen that especially deep learning is used intensively. At this point, studies have revealed different types of models. The number of models and their successes are increasing day by day.

show abstract

“…Secondly, they identified the optimal metric combination, including cosine distance metric, further reducing intra-class differences and thus addressing lighting and viewpoint change issues. On the other hand, Sikdar et al [10] transformed inputs into pyramids of different resolutions, then inputted multiple differently scaled feature maps into the network for learning, and finally adjusted back to the original size before performing max-pooling operations. This processing method has been demonstrated to yield scale-invariant results for re-identification systems.…”

Section: Introductionmentioning

confidence: 99%

Deep learning in person re-identification: a survey

Xiao,

Zhang,

Ning

et al. 2024

International Conference on Image Processing and Artificial Intelligence (ICIPAl 2024)

View full text Add to dashboard Cite

In the era of advancing deep learning, person re-identification has gained widespread application in domains such as video and security monitoring. Person re-identification seeks to recognize and match target persons across images taken by diverse cameras, thereby verifying whether the pedestrian subjects observed by cameras at various locations and times are indeed the same individual. This article categorizes current research into two main types: image-based person reidentification and video-based person re-identification. It provides a thorough overview and analysis of existing literature concerning classification methods, verification models, attention mechanisms, and metric learning,all centered around the research subjects. Additionally, it outlines the progression of datasets and provides insights into the anticipated future trends in person re-identification

show abstract

Scale-invariant batch-adaptive residual learning for person re-identification

Cited by 11 publications

References 5 publications

SiCoDeF² Net: Siamese Convolution Deconvolution Feature Fusion Network for One-Shot Classification

SiCoDeF² Net: Siamese Convolution Deconvolution Feature Fusion Network for One-Shot Classification

Deep Learning Applications for Hyperspectral Imaging: A Systematic Review

Deep learning in person re-identification: a survey

Contact Info

Product

Resources

About