Beyond Part Models: Person Retrieval with Refined Part Pooling

By Yifan Sun, Liang Zheng, Yi Yang, Qi Tian, Shengjin Wang. European Conference on Computer Vision, 2017

2,405 citations recorded by the source corpus.

Employing part-level features for pedestrian image description offers fine-grained information and has been verified as beneficial for person retrieval in very recent literature. A prerequisite of part discovery is that each part should be well located. Instead of using external cues, e.g., pose estimation, to directly locate parts, this paper lays emphasis on the content consistency within each part. Specifically, we target at learning discriminative part-informed features for person retrieval and make two contributions. (i) A network named Part-based Convolutional Baseline (PCB). Given an image input, it outputs a convolutional descriptor consisting of several part-level features. With a uniform partition strategy, PCB achieves competitive results with the state-of-the-art methods, proving itself as a strong convolutional baseline for person retrieval. (ii) A refined part pooling (RPP) method. Uniform partition inevitably incurs outliers in each part, which are in fact more similar to other parts. RPP re-assigns these outliers to the parts they are closest to, resulting in refined parts with enhanced within-part consistency. Experiment confirms that RPP allows PCB to gain another round of performance boost. For instance, on the Market-1501 dataset, we achieve (77.4+4.2)% mAP and (92.3+1.5)% rank-1 accuracy, surpassing the state of the art by a large margin.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/753d2a35c9edf5dfcac4ef3a6adc993b657b01f0

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

Patch-based Knowledge Distillation for Lifelong Person Re-Identification (2022, background): “Person ReID has made remarkable progress under various settings such as supervised learning [18, 40, 57], unsupervised learning [9, 36] and unsupervised domain adaptation [12, 15, 66, 75, 76].”
Hi-CMD: Hierarchical Cross-Modality Disentanglement for Visible-Infrared Person Re-Identification (2019, methodology): “Although the PCB method achieves high performance in single-modality person ReID, a significant performance drop is inevitable in the VI-ReID task.”
Bilinear Attention Networks for Person Retrieval (2019, background): “Recent studies in person re-identification suggest that independent modeling of part regions can enhance the precision of the overall system [36, 38, 20].”
Memory-Based Neighbourhood Embedding for Visual Recognition (2019, methodology): “Following the practice in [36], we adopt ResNet-50 [10] as the backbone network, and change the stride of the last down sampling block from 2 to 1, which is helpful to get more fine-grained features.”
Hierarchical Clustering With Hard-Batch Triplet Loss for Person Re-Identification (2019, methodology): “With the development of convolutional neural networks (CNN), supervised re-ID [9, 11, 21, 25, 30, 2, 31, 23] has achieved excellent performance.”
Learning Compatible Embeddings (2021, background): “Visual search or retrieval systems [36, 37] are widely used in many real-world applications such as face recognition [42, 39, 33, 26, 27], person re-identification [40, 16, 34], car re-identification [19] and image retrieval [2, 12].”
Mixed High-Order Attention Network for Person Re-Identification (2019, methodology): “Implementation: The proposed MHN is applied on both ResNet50-based IDE [63] and PCB [43] architectures.”
Bag of Tricks and a Strong Baseline for Deep Person Re-Identification (2019, neutral): “6 PCB [16] 6 93.”
Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification (2020, positive): “Moreover, we almost approach fully-supervised learning performances (Sun et al., 2018; Ge et al., 2018) without any manual annotations on the target domain.”
Data Augmentation Using Random Image Cropping and Patching for Deep CNNs (2018, methodology): “We employed PCB + RPP model, which is the model of the highest performance in the PCB paper [44].”
Horizontal Pyramid Matching for Person Re-identification (2018, methodology): “The reason why the HPM does not drops dramatically at some point as introduced in (Sun et al. 2017b) is that the pyramid structure can combine both global and local features, which may increase the discriminative ability of very small partition.”
FastReID: A Pytorch Toolbox for General Instance Re-identification (2020, methodology): “1 - PCB [15] (ECCV’18) 92.”
Cross-Modality Person Re-Identification With Shared-Specific Feature Transfer (2020, background): “Some of them treat it as a partial feature learning task and pay much attention to more powerful network structures to better discover, align, and depict the body parts [10, 38, 39, 26].”
Counterfactual Attention Learning for Fine-Grained Visual Categorization and Re-identification (2021, neutral): “We applied the modification backbone due to its competitive performance and relatively concise structure, since recent state-of-the-art methods usually use stronger or more sophisticated backbone such as part model [52, 58] to improve performance.”
AANet: Attribute Attention Network for Person Re-Identifications (2019, methodology): “In the first comparison, which uses ResNet-50, the networks selected are KPM (Res-50) [17], PCB (Res50)[20] and the proposed AANet-50.”
Batch Coherence-Driven Network for Part-Aware Person Re-Identification (2020, methodology): “Methods in this category [2], [14], [16]–[19], [21], [26] perform body part detection and part-level feature extraction sequentially during both the training and testing phases.”
Attack Detection (2020, neutral): “The findings will have significant implications for government agencies, border control systems, and financial institutions, where facial recognition plays a critical role in identity verification and fraud prevention (Sun et al., 2017).”
Attribute Prototype Network for Zero-Shot Learning (2020, methodology): “Our work is related to methods that draw attention over local features [18, 36].”
Each Part Matters: Local Patterns Facilitate Cross-View Geo-Localization (2020, methodology): “Currently, some state-of-the-art works [42], [43], [44], [45] extend the PCB with more partitions or optimization losses.”