Quantized Distillation: Optimizing Driver Activity Recognition Models for Resource-Constrained Environments

Deep learning-based models are at the top of most driver observation benchmarks due to their remarkable accuracies but come with a high computational cost, while the resources are often limited in real-world driving scenarios. This paper presents a lightweight framework for resource- efficient driver activity recognition. We enhance 3D MobileNet, a speed-optimized neural architecture for video classification, with two paradigms for improving the trade-off between model accuracy and computational efficiency: knowledge distillation and model quantization. Knowledge distillation prevents large drops in accuracy when reducing the model size by harvesting knowledge from a large teacher model (I3D) via soft labels instead of using the original ground truth. Quantization further drastically reduces the memory and computation requirements by representing the model weights and activations using lower precision integers. Extensive experiments on a public dataset for in-vehicle monitoring during autonomous driving show that our proposed framework leads to an 3- fold reduction in model size and 1. 4-fold improvement in inference time compared to an already speed-optimized architecture. Our code is available at https://github.com/calvintanama/qd-driver-activity-reco.

Authors

• Calvin Tanama
• Kunyu Peng
• Zdravko Marinov
• Rainer Stiefelhagen
• Alina Roitberg

Keywords

• Solid modeling
• Quantization (signal)
• Three-dimensional displays
• Computational modeling
• Memory management
• Activity recognition
• Computational efficiency
• Action Recognition
• Driver Activity Recognition
• Accuracy Of Model
• Teacher Model
• Model Size
• Model Weights
• Computational Requirements
• Inference Time
• Efficient Recognition
• Accuracy Drop
• Quantification Model
• Deep Learning-based Models
• Fold Reduction
• Soft Labels
• Neural Network
• Deep Learning
• Limited Resources
• Convolutional Neural Network
• Hyperparameters
• Validation Set
• Student Model
• Forward Pass
• Model Inference
• Matrix Multiplication
• Inference Speed
• Resource Efficiency
• Lightweight Architecture
• Ground Truth Labels
• Quantization Scheme
• Depthwise Separable Convolution