A Siamese Long Short-Term Memory Architecture for Human Re-identification

By Rahul Rama Varior, Bing Shuai, Jiwen Lu, Dong Xu, G. Wang. 2016

593 citations recorded by the source corpus.

Matching pedestrians across multiple camera views known as human re-identification (re-identification) is a challenging problem in visual surveillance. In the existing works concentrating on feature extraction, representations are formed locally and independent of other regions. We present a novel siamese Long Short-Term Memory (LSTM) architecture that can process image regions sequentially and enhance the discriminative capability of local feature representation by leveraging contextual information. The feedback connections and internal gating mechanism of the LSTM cells enable our model to memorize the spatial dependencies and selectively propagate relevant contextual information through the network. We demonstrate improved performance compared to the baseline algorithm with no LSTM units and promising results compared to state-of-the-art methods on Market-1501, CUHK03 and VIPeR datasets. Visualization of the internal mechanism of LSTM cells shows meaningful patterns can be learned by our method.

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

Few-Shot Adversarial Domain Adaptation (2017, background): “Siamese networks [7] work well for subspace learning and have worked very well with deep convolutional neural networks [9, 53, 30, 61].”
Spatio-Temporal LSTM with Trust Gates for 3D Human Action Recognition (2016, background): “Recurrent neural networks (RNNs) which are a variant of neural nets for handling sequential data with variable length, have been successfully applied to language modeling [10–12], image captioning [13, 14], video analysis [15–24], human re-identification [25,26], and RGB-based action recognition [27”
Semantics Disentangling for Text-To-Image Generation (2019, methodology): “Inspired by the advantages of Siamese structure used in different tasks [32, 33, 4, 10, 43] which can find the similarity between a pair of sequences, we treat our discriminator as an image comparator so as to preserve the semantic consistency among the generated images as long as their descriptions”
Unified Deep Supervised Domain Adaptation and Generalization (2017, background): “Among algorithms for subspace learning, Siamese networks [11] work well for different tasks [14, 55, 35, 63, 9].”
Person Re-Identification by Camera Correlation Aware Feature Augmentation (2017, methodology): “Nowadays, deep feature learning for person re-id has attracted increasing attention [26,27,29,56,57,58,59,60,61,62].”
DF-GAN: Deep Fusion Generative Adversarial Networks for Text-to-Image Synthesis (2020, methodology): “SD-GAN [15] employs the Siamese structure [48], [49] to distill the semantic commons from texts for image generation consistency.”
Scalable Person Re-identification on Supervised Smoothed Manifold (2017, background): “Thanks to plenty of training images provided, training deep neural networks becomes feasible on this dataset and preferred by most previous works [47, 50, 49].”
Deep Transfer Learning for Person Re-Identification (2016, background): “However, for Re-ID, with a few exceptions [49, 50], this subtraction + binary cross-entropy loss is more popular [51, 57, 48].”
SVDNet for Pedestrian Retrieval (2017, background): “[24] incorporate long short-term memory (LSTM) modules into a siamese network.”
Gated Siamese Convolutional Neural Network Architecture for Human Re-identification (2016, background): “In the recent years, several CNN architectures [1,4,21,43,46,48,52] have been proposed for human re-identification.”
Knowledge Distillation for End-to-End Person Search (2019, unknown): “More recent re-identiﬁcation approaches are based on CNNs [1, 31, 62, 64] and mainly concerned with the estimation of a ID-feature embedding spaces, either via Siamese networks and contrastive losses [1, 9, 11, 15, 31, 36, 50, 56, 62], or with ID-classiﬁcation and cross-entropy losses [54, 67].”
Pose-Guided Feature Alignment for Occluded Person Re-Identification (2019, background): “Recent years, deep learning methods [39, 34, 13, 27, 3, 25, 7, 6, 36, 21] for person re-id show signiﬁcant superiority on retrieval accuracy.”
HydraPlus-Net: Attentive Deep Features for Pedestrian Analysis (2017, methodology): “The Market-1501 [38] dataset is also evaluated with the metric learning WARCA-L [11], a novel Siamese LSTM architecture LOMO+CN [27], the Siamese CNN with learnable gate S-CNN [26], and the bag of words model BoW-best [38].”
Pose-Driven Deep Convolutional Model for Person Re-identification (2017, methodology): “Predefined rigid body parts are also used by many deep learning based methods [5, 47, 40] for the purpose of learning local pedestrian features.”
Part-Aligned Bilinear Representations for Person Re-identification (2018, background): “The approaches in this stream of research represent an image as a combination of local descriptors, where each lo-cal descriptor represents a spatial partition such as grid cell [8–10] and horizontal stripe [11–13].”
Learning Deep Context-Aware Features over Body and Latent Parts for Person Re-identification (2017, methodology): “Market1501: For the Market1501 dataset, several state-of-the-art methods are compared, including Bag of Words (BOW) [53], Weighted Approximate Rank Component Analysis (WARCA) [15], Discriminative Null Space (DNS) [47], Spatially Constrained Similarity function on Polynomial feature map (SCSP) [2], a”
Attribute-Aware Attention Model for Fine-grained Representation Learning (2018, methodology): “Deep learning methods [1, 36, 46] learn image representations and distance metrics simultaneously in an end-to-end manner.”