Bag of Tricks and a Strong Baseline for Deep Person Re-Identification

By Hao Luo, Youzhi Gu, Xingyu Liao, Shenqi Lai, Wei Jiang. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019

1,461 citations recorded by the source corpus.

This paper explores a simple and efficient baseline for person re-identification (ReID). Person re-identification (ReID) with deep neural networks has made progress and achieved high performance in recent years. However, many state-of-the-arts methods design complex network structure and concatenate multi-branch features. In the literature, some effective training tricks are briefly appeared in several papers or source codes. This paper will collect and evaluate these effective training tricks in person ReID. By combining these tricks together, the model achieves 94.5% rank-1 and 85.9% mAP on Market1501 with only using global features. Our codes and models are available at https://github.com/michuanhaohao/reid-strong-baseline.

Subfields: Computer Vision and Pattern Recognition, Ocean Engineering

View Paper: https://www.semanticscholar.org/paper/1e7ff66f484e9c7b235b4335103619c8796af419

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

In Ocean Engineering

Citations in the corpus

CLIP-ReID: Exploiting Vision-Language Model for Image Re-Identification without Concrete Text Labels (2022, methodology): “2016) has been regarded as the common model (Luo et al. 2019) for most ReID datasets.”
Learning Continual Compatible Representation for Re-indexing Free Lifelong Person Re-identification (2024, methodology): “Considering that the RFL-ReID task aims to continuously match the same person across the sequentially given datasets, we employ a cross-entropy loss L ce [13] and a triplet loss L trip [8] to learn a discriminative representation for each person in the current dataset D s .”
RGB-Infrared Cross-Modality Person Re-Identification via Joint Pixel and Feature Alignment (2019, methodology): “Existing methods can be grouped into hand-crafted descriptors [22, 36, 19, 34], metric learning methods [43, 16, 20] and deep learning algorithms [42, 11, 28, 21, 44, 30, 15, 35, 33, 18, 17, 1, 31].”
MaxUp: A Simple Way to Improve Generalization of Neural Network Training (2020, neutral): “This makes OHEM particularly useful for class-imbalance tasks, e.g. object detection (Shrivastava et al., 2016), person re-identiﬁcation (Luo et al., 2019).”
GaitPart: Temporal Part-Based Model for Gait Recognition (2020, methodology): “the similar operations are commonly used in [5, 6, 20, 15].”
Patch-based Knowledge Distillation for Lifelong Person Re-Identification (2022, methodology): “Note that the last stride is set to 1 as suggested in [40].”
Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification (2020, methodology): “The classification and triplet losses are commonly adopted together to achieve state-of-the-art performances in both fully-supervised (Luo et al., 2019) and unsupervised (Zhang et al.”
Continual Representation Learning for Biometric Identification (2020, methodology): “We use the popular person re-id testing protocol [40].”
A Strong Baseline and Batch Normalization Neck for Deep Person Re-Identification (2019, background): “All hyper-parameters are set same with [41] and [1].”
Identity-Guided Human Semantic Parsing for Person Re-Identification (2020, methodology): “For each basic loss group, we follow [28] to combine the triplet loss [10] and cross-entropy loss with label smoothing [39].”
Self-paced Contrastive Learning with Hybrid Memory for Domain Adaptive Object Re-ID (2020, methodology): “7 Supervised learning methods on source DukeMTMC-reID Market-1501 Bags of tricks [28] CVPRW’19 76.”
Diverse Part Discovery: Occluded Person Re-identification with Part-Aware Transformer (2021, methodology): “For classifiers, as in [28] we use a batch normalization layer [17] and a fully connected layer followed by a softmax function.”
Cross-Modality Person Re-Identification With Shared-Specific Feature Transfer (2020, methodology): “The augmentations include: (1) the feature blocks are all set to BNNeck [30]; (2) the input images are augmented with random erasing [54].”
Diverse Knowledge Distillation for End-to-End Person Search (2020, methodology): “As shown in Table 6, when removing the important schemes in the Re-ID model, i.e. , random erasing (Zhong et al. 2020) and BNNeck (Luo et al. 2019), the performance is slightly lower than the full model yet outperforms all the recent methods.”
Style Normalization and Restitution for Generalizable Person Re-Identification (2020, methodology): “Taking the widely used ReID network of ResNet-50 [10, 1, 28] as an example (see Figure 2(a)), SNR module is added after each convolutional block.”
TransReID: Transformer-based Object Re-Identification (2021, methodology): “[23] proposed the BNNeck to better combine ID loss and triplet loss.”