Основні поняття
Proposing a hybrid model combining self-attention and BiLSTM approaches significantly improves age and gender classification accuracy.
Анотація
Abstract:
Computer vision advancements lead to new applications like visual surveillance, targeted ads, etc.
Face analysis crucial; age and gender classification challenging.
Proposed hybrid model combines self-attention and BiLSTM for improved accuracy.
Introduction:
Face features used in various domains.
Deep learning revolutionized image processing.
Transfer Learning addresses data availability issues.
Dataset:
Adience face dataset used for training/testing.
Pre-processing steps involved data cleaning, face detection, and normalization.
Model Architecture:
Proposed model combines ViT's self-attention with BiLSTM (h-Sequencer).
Detailed description of the proposed model's components provided.
Experimental Setup:
Nvidia A5000 RTX 24GB GPU used for experiments.
Models evaluated using 5-fold cross-validation with/without data augmentation.
Results:
Proposed model outperforms other models in both age and gender classification tasks.
Achieves approximately 10% improvement over state-of-the-art implementations.
Conclusion:
Hybrid model shows superior performance in age and gender classification tasks.
Статистика
An improvement of approximately 10% and 6% over the state-of-the-art implementations for age and gender classification, respectively, are noted for the proposed model.