Authors: Andrey Davydov, Alexey Sidnev, Artsiom Sanakoyeu, Yuhua Chen, Mathieu Salzmann, Pascal Fua

Published on: February 05, 2024

Impact Score: 8.07

Arxiv code: Arxiv:2402.02736

Summary

• What is new: The novel approach uses unannotated videos to improve human body pose and shape estimation in scenarios lacking sufficient annotated data.
• Why this is important: The challenge of accurately estimating human body pose and shape with limited annotated training data.
• What the research proposes: A method leveraging unannotated videos to enrich training data by enforcing consistency between optical flow and pose changes across video frames.
• Results: With this approach, the performance matches methods that relied on significantly more annotated data.

Technical Details

Technological frameworks used: Supervised deep learning with added unsupervised data consistency checks

Models used: Deep-learning algorithms for human pose and shape estimation

Data used: Unannotated videos and limited available annotated data

Potential Impact

Fitness and sports analytics, security and surveillance, animation and gaming industries, wearable tech companies

Want to implement this idea in a business?

We have generated a startup concept here: FlowFitAI.