Person re-identification is the task of associating images of the same person taken from different cameras or from the same camera in different occasions, the process of person re-identification is shown in figure below, taken from this paper survey.
This repository contains the code of implementation person re-identification using triplet loss and SVM classifier with our custom dataset. The dataset which we are olympic swimmer dataset from YouTube in the top view angle, you can use your own dataset as well.- Tensorflow 2.3
- tensorflow-addons
- numpy
- imbalanced-learn (incase the data is imbalanced)
- scikit-learn 0.23.1
Metric | Validation Loss | Validation Accuracy |
---|---|---|
CNN (Cross entropy) | 0.7284 | 80% |
CNN (Triplet Loss) + SVM | 0.0014 | 99% |