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

Table 2

Size and accuracy of pruned network based on proposed method.

Pruning ratio	0.5	0.6	0.7	0.8	0.9
Size of EX (KB)	171	143	117	92	69
Accuracy of EX	90.8%	90.9%	90.9%	88.9%	88.7%
Size of NEX (KB)	101	71	49	33	19
Accuracy of NEX	86.3%	90.0%	89.8%	90.7%	89.1%