Model Lightweighting for Real-time Distraction Detection on Resource-Limited Devices

Algorithm 1

The proposed CNN light-weighting method.

	Input: training data , validation data ;
	the th convolution layer with filters ;
	the th convolution layer bottleneck residual block number ;
	the th convolution layer channel number ;
	Output: the light-weighting model and the updated , , and .
(1)	for in convolution layers do
(2)	Optimization n and c based on (2);
(3)	Fine-tuning the module;
(4)	Update n and c based on accuracy;
(5)	for pruning ratios ⟵ 0.1 to 0.9 do
(6)	Fine-tuning the module;
(7)	Update pruning ratios based on accuracy controlled decrease within ;
(8)	end for
(9)	end for
(10)	Get final parameters , , and and fine-tuning the pruned model with .