Receptive Multi-Granularity Representation for Person Re-Identification

Abstract

A key for person re-identification is achieving consistent local details for discriminative representation across variable environments. Current stripe-based feature learning approaches have delivered impressive accuracy, but do not make a proper trade-off between diversity, locality, and robustness, which easily suffers from part semantic inconsistency for the conflict between rigid partition and misalignment. This paper proposes a receptive multi-granularity learning approach to facilitate stripe-based feature learning. This approach performs local partition on the intermediate representations to operate receptive region ranges, rather than current approaches on input images or output features, thus can enhance the representation of locality while remaining proper local association. Toward this end, the local partitions are adaptively pooled by using significance-balanced activations for uniform stripes. Random shifting augmentation is further introduced for a higher variance of person appearing regions within bounding boxes to ease misalignment. By twobranch network architecture, different scales of discriminative identity representation can be learned. In this way, our model can provide a more comprehensive and efficient feature representation without larger model storage costs. Extensive experiments on intra-dataset and cross-dataset evaluations demonstrate the effectiveness of the proposed approach. Especially, our approach achieves a state-of-the-art accuracy of 96.2%@Rank-1 or 90.0%@mAP on the challenging Market-1501 benchmark.

Focus on the Visible Regions: Semantic-Guided Alignment Model for Occluded Person Re-Identification

“…According to [ 18 ], combining multi-granularity information can obtain more discriminative features, so the model uses five features for training, including four local features and one global feature. Five feature vectors are fed into a 1 × 1 convolutional layer to reduce the dimension of vectors from 2048 to 256.…”

“…In recent years, pedestrian re-identification methods based on deep learning have significantly improved retrieval accuracy. Recent works [ 2 , 17 , 18 ] show that combining local features of body parts can construct a more efficient representation. For example, Sun et al [ 2 ] split the feature map uniformly and used multiple classifiers to learn part-level features.…”

“…Wei et al [ 17 ] employed pose landmarks to extract local features of pedestrians and to connect these features for retrieval. Wang et al [ 18 ] partitioned the images into several stripes, and the model learned fine-grained features and combined global features to form multi-scale descriptors.…”

A Discriminative Person Re-Identification Model With Global-Local Attention and Adaptive Weighted Rank List Loss

“…They trained discriminant models to extract features and used similarity measures to determine the similarity between images. Wang et al [8] designed a multigranularity network (MGN) that combines part multigranularity information and global information to extract rich visual features. Zheng et al [10] combined discriminant and generative learning to train powerful feature extractors.…”

“…The methods of MGN [8], CAM [9], GD-Net [10], MSBA [12], BagTricks [14], AGW [15], ABD-Net [16], RAG-SC [17], AANet [22], SONA [24], RRGCCAN [26], IANet [29], EMM [32], MPM-LTL [35], HA-CNN [46] and SCAL [47] are selected for comparison. It is worth noting that we do not use the re-ranking strategy.…”