Pose-Invariant Embedding for Deep Person Re-Identification

By Liang Zheng, Yujia Huang, Huchuan Lu, Yi Yang. IEEE Transactions on Image Processing, 2017

593 citations recorded by the source corpus.

Pedestrian misalignment, which mainly arises from detector errors and pose variations, is a critical problem for a robust person re-identification (re-ID) system. With poor alignment, the feature learning and matching process might be largely compromised. To address this problem, this paper introduces pose-invariant embedding (PIE) as a pedestrian descriptor. First, in order to align pedestrians to a standard pose, the PoseBox structure is introduced, which is generated through pose estimation followed by affine transformations. Second, to reduce the impact of pose estimation errors and information loss during the PoseBox construction, we design a PoseBox fusion (PBF) CNN architecture that takes the original image, the PoseBox, and the pose estimation confidence as input. The proposed PIE descriptor is thus defined as the fully connected layer of the PBF network for the retrieval task. Experiments are conducted on the Market-1501, CUHK03-NP, and DukeMTMC-reID datasets. We show that PoseBox alone yields decent re-ID accuracy and that when integrated in the PBF network, the learned PIE descriptor produces competitive performance compared with state-of-the-art approaches.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/1aaf1c043e734321ddbdceb0392fd89c062fb33a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

1. Liang Zheng
2. Yi Yang
3. Zhedong Zheng
4. Qi Tian
5. Shiliang Zhang
6. Wanli Ouyang
7. Wen Gao
8. Shengjin Wang

In Computer Vision and Pattern Recognition

1. Yi Yang
2. Dacheng Tao
3. Liang Zheng
4. Xiaogang Wang
5. Luc Van Gool
6. Qi Tian
7. Ming-Hsuan Yang
8. Wanli Ouyang

Citations in the corpus

• Human Semantic Parsing for Person Re-identification (2018, background): “Some works [47, 34] have tried to address the misalignment issue by explicitly integrating pose estimation into person re-identification where off-the-shelf pose estimation models are used to initialize part locations as quadrilaterals which then are aligned via”
• Part-Aligned Bilinear Representations for Person Re-identification (2018, methodology): “The complexity of the R-CPM is lower than the standard FCN ( ∼ 6M param) used in [14] and CPM ( ∼ 30M param) used in [15].”
• SVDNet for Pedestrian Retrieval (2017, methodology): “So a popular deep learning method for re-ID consists of 1) training a classification deep model on the training set, 2) extracting image descriptors using the fullyconnected (FC) layer for the query and gallery images, and 3) computing similarities based on Euclidean distance before returning the so”
• Pose-Driven Deep Convolutional Model for Person Re-identification (2017, methodology): “Different from these algorithms, our work and the ones in [62, 16] use more accurate human pose estimation algorithms to acquire human pose features.”
• AlignedReID: Surpassing Human-Level Performance in Person Re-Identification (2017, background): “Other works use pose estimation result for the alignment [52, 37, 50], which requires additional supervision and a pose estimation step (which is often error-prone).”
• Beyond Part Models: Person Retrieval with Refined Part Pooling (2017, methodology): “Several recent works in re-ID employ these tools for pedestrian partition and report encouraging improvement [38, 27, 31].”
• Deep Representation Learning With Part Loss for Person Re-Identification (2017, methodology): “Inspired by the success of Convolutional Neural Network (CNN) in large-scale visual classification [20], lots of approaches have been proposed to generate representations based on CNN [19], [21]–[30].”
• Learning Feature Pyramids for Human Pose Estimation (2017, background): “Localizing body parts for human body is a fundamental yet challenging task in computer vision, and it serves as an important basis for high-level vision tasks, e.g ., activity recognition [60, 54], clothing parsing [57, 58, 36], human re-identification [65], and human-computer interaction.”
• Mask-Guided Contrastive Attention Model for Person Re-identification (2018, methodology): “Inspired by recent progress in pose estimation [13, 3], several pose based person ReID methods are proposed [44, 32, 23, 49].”
• Each Part Matters: Local Patterns Facilitate Cross-View Geo-Localization (2020, background): “aspects: Different from works of the first line [35], [34], [46], [47], [48], the proposed method does not need an extra part estimator.”
• Batch Coherence-Driven Network for Part-Aware Person Re-Identification (2020, background): “caused by variations in viewpoint and human pose [23], [24], as well as errors in pedestrian detection [25].”
• ABD-Net: Attentive but Diverse Person Re-Identification (2019, background): “Many part-based methods have achieved superior performance [2, 3, 26, 27, 28, 4, 5, 29, 6, 30, 31, 8, 32].”
• Diverse Part Discovery: Occluded Person Re-identification with Part-Aware Transformer (2021, methodology): “Existing Re-ID methods can be summarized to hand-crafted descriptors [47, 23], metric learning methods [56, 20, 24] and deep learning methods[38, 25, 33, 35, 45, 52, 19, 21, 34, 26, 22].”
• A Strong Baseline and Batch Normalization Neck for Deep Person Re-Identification (2019, methodology): “Pose-guided methods [18]–[21] use an extra pose/skeleton estimation model to acquire human pose information.”