notesum.ai

Published at November 19

Puppet-CNN: Input-Adaptive Convolutional Neural Networks with Model Compression using Ordinary Differential Equation

cs.AI

Released Date: November 19, 2024

Authors: Yucheng Xing¹, Xin Wang¹

Aff.: ¹Department of Electrical and Computer Engineering, Stony Brook University, Stony Brook, NY 11794, USA

Arxiv: http://arxiv.org/abs/2411.12876v1

Refer to caption

		Top-1 Acc ( $\uparrow$ )	Top-5 Acc ( $\uparrow$ )	Params Size ( $\downarrow$ )	Inference Speed ( $\downarrow$ )
Adaptive-Params	DFN	68.59 $\%$	93.13 $\%$	75.89MB	0.0012s/img
	WeightNet	62.77 $\%$	93.52 $\%$	45.87MB	0.0040s/img
	DDFN-SW	55.55 $\%$	93.34 $\%$	300.83MB	0.0059s/img
Adaptive-Depth	BranchyNet	70.00 $\%$	93.94 $\%$	27.69MB	0.0015s/img
	SkipNet	55.82 $\%$	87.95 $\%$	68.88MB	0.0115s/img
	DRNN	41.71 $\%$	63.89 $\%$	45.78MB	0.0025s/img
Ours	Puppet-CNN	72.51 $\%$	96.85 $\%$	1.08MB	0.0039s/img