A Discriminatively Learned CNN Embedding for Person Reidentification
By Zhedong Zheng, Liang Zheng, Yi Yang. ACM Trans. Multim. Comput. Commun. Appl., 2016
949 citations recorded by the source corpus.
In this article, we revisit two popular convolutional neural networks in person re-identification (re-ID): verification and identification models. The two models have their respective advantages and limitations due to different loss functions. Here, we shed light on how to combine the two models to learn more discriminative pedestrian descriptors. Specifically, we propose a Siamese network that simultaneously computes the identification loss and verification loss. Given a pair of training images, the network predicts the identities of the two input images and whether they belong to the same identity. Our network learns a discriminative embedding and a similarity measurement at the same time, thus taking full usage of the re-ID annotations. Our method can be easily applied on different pretrained networks. Albeit simple, the learned embedding improves the state-of-the-art performance on two public person re-ID benchmarks. Further, we show that our architecture can also be applied to image retrieval. The code is available at https://github.com/layumi/2016_person_re-ID.
Subfields: Computer Vision and Pattern Recognition
View Paper: https://www.semanticscholar.org/paper/193089d56758ab88391d846edd08d359b1f9a863
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
- University-1652: A Multi-view Multi-source Benchmark for Drone-based Geo-localization (2020, methodology): “We also evaluate three basic models and three different loss terms, including contrastive loss [16, 35, 43], triplet loss [5, 6], and instance loss [42].”
- In Defense of the Triplet Loss for Person Re-Identification (2017, background): “Indeed, pretrained models often obtain great scores for person ReID [10, 52], while ever fewer top-performing approaches use networks trained from scratch [21, 1, 8, 42, 31, 39, 4].”
- TransReID: Transformer-based Object Re-Identification (2021, methodology): “[47] proposed the ID-discriminative embedding (IDE) to train the ReID model as image classification and it is fine-tuned from the ImageNet [6] pre-trained models.”
- A Bottom-Up Clustering Approach to Unsupervised Person Re-Identification (2019, methodology): “Recently, deep learning methods achieve great success (Li et al. 2014; Varior, Haloi, and Wang 2016; Zheng et al. 2015; Zheng, Zheng, and Yang 2017) by simultaneously learning the image representations and similarities.”
- Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in Vitro (2017, methodology): “Our baseline results are on par with the those reported in [50, 52].”
- Deep Representation Learning With Part Loss for Person Re-Identification (2017, methodology): “We first evaluate the performance of our baseline network and compare it with CNN networks used in existing methods [22], [25].”
- Improving Person Re-identification by Attribute and Identity Learning (2017, methodology): “In [45, 8], the combination of verification and classification losses is proven effective, consistent with the findings in [29].”
- Person Transfer GAN to Bridge Domain Gap for Person Re-identification (2017, methodology): “Some other works [39, 6] combine verification models with classification models to learn descriptors.”
- Pedestrian Alignment Network for Large-scale Person Re-Identification (2017, methodology): “The learned embeddings extracted from the fine-tuned CNNs are shown to outperform the hand-crafted features [Zheng et al., 2016b,Xiao et al., 2016,Zhong et al., 2017].”
- Pose Transferrable Person Re-identification (2018, methodology): “task, as well as design of novel discriminative loss functions suited for ReID task such as contrastive loss [49] and triplet loss [21].”
- A Variational U-Net for Conditional Appearance and Shape Generation (2018, result): “These results are comparable to those reported in [48].”
- SVDNet for Pedestrian Retrieval (2017, background): “Both works [10] and [34] combine the verification and classification losses, and we will investigate into integrating this strategy into SVDNet.”
- A Strong Baseline and Batch Normalization Neck for Deep Person Re-Identification (2019, methodology): “However, few works [5], [8], [9] have focused on the design of an effective baseline.”
- MOTS: Multi-Object Tracking and Segmentation (2019, methodology): “The association head is inspired by the embedding vectors used in person re-identification [17, 5, 31, 55, 62].”