Conceitos Básicos
Efficiently estimate identity, gender, and age from hand images for criminal investigations.
Resumo
The paper proposes a multi-task representation learning framework to jointly estimate the identity, gender, and age of individuals from hand images. It explores deep learning architectures for this purpose and evaluates their performance on a dataset of 11k hands. The study aims to assist international police forces in identifying and convicting abusers based on hand image analysis.
- Introduction to biometric identification using body parts.
- Importance of hand images in criminal investigations.
- Comparison of different deep learning architectures for joint estimation.
- Description of the proposed multi-task representation learning framework.
- Details on network architecture and loss function.
- Performance evaluation with convolution-based and transformer-based architectures.
- Results showing the effectiveness of the proposed method.
- Conclusion highlighting the significance of the research.
Estatísticas
We make extensive evaluations and comparisons on a publicly available 11k hands dataset.
The ViT-B-16 has almost 3 times more parameters than Swin-T i.e., 86.6M vs 28.3M.
Citações
"Hand images are often the only available information in cases of serious crime such as sexual abuse."
"Our proposed deep multi-task representation learning framework jointly estimates the identity, gender, and age of individuals from their hand images."