Visual Interaction Force Estimation Based on Time-Sensitive Dual-Resolution Learning Network

Table 1

Comparison results with state-of-the-art force estimation methods.

ID	Method	RMSE	MAE	MSE		Inference time (s)
(a)	VGG19 (, 400000 images)	0.2201	0.1969	0.0484	-0.0279
(b)	Resnet (, 400000 images)	0.2197	0.1914	0.0483	-0.0215
(c)	VGG+LSTM (, 400000 images)	0.2161	0.1825	0.0467	-0.0026
(d)	Resnet+LSTM (, 400000 images)	0.2182	0.1840	0.0476	-0.1165
(e)	3DCNN+Attention+LSTM	0.0955	0.0312	0.0091	—
(f)	Our method (, 40000 images)	0.0480	0.0329	0.0023	0.9718
(g)	Our method (, 400000 images)	0.0397	0.0243	0.0015	0.9725
(h)	Simplified version of our method (, 400000 images)	0.0313	0.0183	0.0009	0.9833
(i)	Simplified version of our method (, 400000 images)	0.0295	0.0185	0.0008	0.9850