Small Network for Lightweight Task in Computer Vision: A Pruning Method Based on Feature Representation

Table 1

Detail of experiments.

No.	Original network	Pretraining	Refining	Superparameter	Pruning method
1	Cifar10_full	Cifar10	Cifar10	160 epoch 0.001lr	Map-EX, Size-EX, scratch-EX, Map-NEX, Size-NEX, and scratch-NEX
			Modified Cifar10
			Modified Cifar10 with different data number
			Cifar10	160 epoch 0.001lr	Multisample pruning with different ratio
2	VGG16	Cifar10	Cifar10	160 epoch 0.001lr	Map-NEX, multisample pruning (0.8), and network slimming
			ImageNet subdataset
3	YoloV3	ImageNet	VOC	80 epoch 0.001lr; 80 epoch 0.001lr	Map-NEX; Size-NEX