TransReID: Transformer-based Object Re-Identification

He, Shuting; Luo, Hao; Wang, Pichao; Wang, Fan; Li, Hao; Jiang, Wei

doi:10.1109/iccv48922.2021.01474

Cited by 672 publications

(328 citation statements)

References 39 publications

Supporting

Mentioning

324

Contrasting

Order By: Relevance

“…TTSR [Yang et al, 2020] restored the texture information of the image super-resolution result based on the transformer. TransReID [He et al, 2021] applied the transformer to the field of retrieval for the first time and achieved similar results with the CNN-based method.…”

Section: Transformer In Visionmentioning

confidence: 83%

“…In addition, to express the global features we embed an additional embedding patch for the global feature representation. After passing through the Transformer Layer (The VitS used in this paper consists of a total of 8 self-attentive blocks, with the same structure as TransReid [He et al, 2021]), we get the features with the same dimension as the input, and the cls token is regarded as the final feature representation. The output is divided into two parts, one of which converts the 768-dimensional VitS output into a 512-dimensional feature vector via a fully-connected layer to calculate Triplet Loss, and the other part goes through a fully-connected layer on this 512-dimensional feature vector for classification.…”

Section: Benchmark Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Vision-Based UAV Localization System in Denial Environments

Dai¹,

Huang²,

Zhuang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Unmanned Aerial Vehicle (UAV) localization capability is critical in a Global Navigation Satellite System (GNSS) denial environment. The aim of this paper is to investigate the problem of locating the UAV itself through a purely visual approach. This task mainly refers to: matching the corresponding geo-tagged satellite images through the images acquired by the camera when the UAV does not acquire GNSS signals, where the satellite images are the bridge between the UAV images and the location information. However, the sampling points of previous cross-view datasets based on UAVs are discrete in spatial distribution and the inter-class relationships are not established. In the actual process of UAV-localization, the inter-class feature similarity of the proximity position distribution should be small due to the continuity of UAV movement in space. In view of this, this paper have reformulated an intensive dataset for UAV positioning tasks, which is named DenseUAV, aiming to solve the problems caused by spatial distance and scale transformation in practical application scenarios, so as to achieve high-precision UAV-localization in GNSS denial environment. In addition, a new continuum-type evaluation metric named SDM is designed to evaluate the accuracy of model matching by exploiting the continuum of UAVs in space. Specifically, with the ideas of siamese networks and metric learning, a transformer-based baseline was constructed to enhance the capture of spatially subtle features. Ultimately, a neighbor-search post-processing strategy was proposed to solve the problem of large distance localisation bias.

show abstract

Section: Transformer In Visionmentioning

confidence: 83%

Section: Benchmark Introductionmentioning

confidence: 99%

Vision-Based UAV Localization System in Denial Environments

Dai¹,

Huang²,

Zhuang³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Some works learn the relationship between body parts. For example, Su et al [32] align image patches that contain the same body part; Xia et al [2] utilize Non-local module to explore the importance of different parts; He et al [16] adopt the Vision Transformer [9] for image-based ReID feature extraction. These methods show excellent performance on public datasets, such as Market-1501 [50] and MSMT17 [37].…”

Section: Person Re-identificationmentioning

confidence: 99%

Spatiotemporal Transformer for Video-based Person Re-identification

Zhang¹,

Xie²,

Zhuang³

et al. 2021

Preprint

View full text Add to dashboard Cite

Recently, the Transformer module has been transplanted from natural language processing to computer vision. This paper applies the Transformer to video-based person reidentification, where the key issue is to extract the discriminative information from a tracklet. We show that, despite the strong learning ability, the vanilla Transformer suffers from an increased risk of over-fitting, arguably due to a large number of attention parameters and insufficient training data. To solve this problem, we propose a novel pipeline where the model is pre-trained on a set of synthesized video data and then transferred to the downstream domains with the perception-constrained Spatiotemporal Transformer (STT) module and Global Transformer (GT) module. The derived algorithm achieves significant accuracy gain on three popular video-based person reidentification benchmarks, MARS, DukeMTMC-VideoReID, and LS-VID, especially when the training and testing data are from different domains. More importantly, our research sheds light on the application of the Transformer on highlystructured visual data.

show abstract

“…It begins with the NPO (non-pedestrian occlusion) augmentation strategy that produces image pairs and occlusion masks. Following [13], we simply adopt Vision Transformer (ViT) [4] as the feature extractor. Position embeddings and a classification [cls] token are prepended to the input image.…”

Section: Feature Erasing and Diffusion Networkmentioning

confidence: 99%