Inferring Body Measurements from 2D Images: A Comprehensive Review.

Area of Science:

• Computer Vision
• Medical Imaging
• Biometrics

Background:

• Estimating anthropometric measurements from 2D images, especially for children, is an under-explored research area.
• Existing research has focused more on pose estimation and body shape classification than on precise measurement prediction.

Purpose of the Study:

• To review current methodologies for predicting anthropometric measurements from 2D facial and full-body images.
• To identify challenges and opportunities in using deep learning and machine learning for body measurement prediction.
• To propose future research directions for improving accuracy and inclusivity.

Main Methods:

• Comprehensive literature review of deep learning and traditional machine learning techniques.
• Analysis of commonly used datasets and their limitations.
• Evaluation of vision transformers and model explainability.

Main Results:

• Deep learning models, particularly when integrated with traditional machine learning, yield the most accurate predictions.
• Current datasets lack diversity, hindering model generalizability.
• Vision transformers show potential for advancing the field.

Conclusions:

• Accurate prediction of anthropometric measurements from images requires further research into inclusive datasets and explainable AI.
• Hybrid deep learning and machine learning approaches are most effective.
• Future work should focus on improving model accuracy and addressing data bias.