Authors: Xiangqun Zhang, Ruize Han, Wei Feng

Published on: February 03, 2024

Impact Score: 8.3

Arxiv code: Arxiv:2402.02108

Summary

• What is new: Using synthetic video data for training in cross-domain person re-identification tasks instead of relying on real-world data.
• Why this is important: Reducing dependency on real training data collection and annotation for video-based person re-identification.
• What the research proposes: A self-supervised domain invariant feature learning strategy combined with a mean-teacher scheme for improving identification in the target domain.
• Results: The method effectively adapted between synthetic and real domains, with synthetic data outperforming real data in cross-domain tasks.

Technical Details

Technological frameworks used: Self-supervised learning, Mean-teacher scheme

Models used: Domain invariant feature learning models

Data used: Synthetic video datasets for training, real-world video datasets for testing

Potential Impact

Security companies, surveillance systems, retail analytics, and any industry relying on video-based identification could benefit or need to adapt.

Want to implement this idea in a business?

We have generated a startup concept here: SynthIdent.