A method and system for unsupervised domain adaptive person re-identification based on density clustering
A density clustering and unsupervised technology, applied in the field of image processing, can solve problems such as no clustering of correct samples, difficulty in adapting to data sets, etc., and achieve the effect of improving discrimination ability
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0069] combine figure 1 As shown, the person re-identification method based on unsupervised domain adaptation based on density clustering includes five parts: supervised learning, feature dynamic storage, adaptive dynamic clustering, cross-camera similarity evaluation and loss optimization. In this embodiment, it is known that: labeled source data in and represent the i1th training sample and its identity label respectively, i1∈[1,N s ], N s is the number of samples in the source domain dataset. is the unlabeled target data, N t is the total number of samples in the target domain dataset, Indicates the i2th training sample in the target domain, i2∈[1,N t ], and represent the selected image, respectively and The feature map output before the last fully connected layer of the selected backbone network, the present invention uses the ResNet-50 model as the benchmark model
[0070] The steps are described below:
[0071] Step 1. Supervised learning:
[0072] I...
Embodiment 2
[0125] As another embodiment of the present invention, a person re-identification system based on density clustering-based unsupervised domain adaptation is provided, including a feature memory, an adaptive dynamic clustering module, a cross-camera similarity evaluation and loss optimization module.
[0126] The feature memory is used to dynamically store features, and store the source domain centroids and target data instances in sequence according to the known identity of the source domain and the target domain index; the feature vector corresponding to the source domain is updated according to the centroid of the source domain sample category, The eigenvector corresponding to the target domain is updated according to the eigenvalue of the target domain sample;
[0127] The adaptive dynamic clustering module is used to dynamically update the clustering radius of the DBSCAN clustering algorithm, first obtain a stable distance measure in the target domain by means of a feature ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More