PHD Discussions Logo

Ask, Learn and Accelerate in your PhD Research

Question Icon Post Your Answer

Question Icon

2 years ago in Machine Learning By Rachna M

Which Machine Learning models are suitable for Person Re-Identification?

Which types of machine learning models are commonly used to solve person re-identification problems?

All Answers (1 Answers In All)

By Usha K Answered 8 months ago

Person Re-Identification is primarily addressed with Deep Learning models due to their superior feature learning.
Metric Learning Models: Use Siamese Networks or Triplet Networks with a contrastive or triplet loss to learn a feature space where same-person images are closer.
Representation Learning Models: Employ deep Convolutional Neural Networks (CNNs) like ResNet50 as backbone feature extractors, often trained with a combined identity (softmax) and metric loss.
Advanced Architectures: Include part-based models (PCB), attention mechanisms, and transformer-based models (ViT). Training requires large Re-ID datasets (Market-1501, DukeMTMC). Start with a ResNet backbone + triplet loss as a strong baseline.

Your Answer