notesum.ai

Published at December 4

Benchmarking Pretrained Attention-based Models for Real-Time Recognition in Robot-Assisted Esophagectomy

cs.CV

cs.AI

Released Date: December 4, 2024

Authors: Ronald L. P. D. de Jong, Yasmina al Khalil¹, Tim J. M. Jaspers, Romy C. van Jaarsveld¹, Gino M. Kuiper¹, Yiping Li¹, Richard van Hillegersberg¹, Jelle P. Ruurda¹, Marcel Breeuwer², Fons van der Sommen²

Aff.: ¹University Medical Center Utrecht; ²Eindhoven University of Technology

Arxiv: http://arxiv.org/pdf/2412.03401v1

		RAMIE dataset		CholecSeg8k dataset
Model	Pretraining	Dice score	ASSD [pixels]	Dice score	ASSD [pixels]
DeepLabv3	No pretraining	0.43 $\pm$ 0.14	38 $\pm$ 26	0.53 $\pm$ 0.09	44 $\pm$ 16
	ImageNet	0.52 $\pm$ 0.16	29 $\pm$ 21	0.58 $\pm$ 0.11	34 $\pm$ 14
	ADE20k	^∗0.56 $\pm$ 0.14^∗	^∗23 $\pm$ 18^∗	^∗0.61 $\pm$ 0.11^∗	^∗30 $\pm$ 13^∗
DeepLabv3+	No pretraining	0.39 $\pm$ 0.14	44 $\pm$ 29	0.50 $\pm$ 0.09	42 $\pm$ 15
	ImageNet	0.51 $\pm$ 0.16	29 $\pm$ 23	0.56 $\pm$ 0.11	36 $\pm$ 13
	ADE20k	^∗0.57 $\pm$ 0.15^∗	^∗24 $\pm$ 21^∗	^∗0.60 $\pm$ 0.11^∗	^∗32 $\pm$ 13^∗
PSPNet	No pretraining	0.38 $\pm$ 0.13	45 $\pm$ 30	0.52 $\pm$ 0.10	38 $\pm$ 15
	ImageNet	0.53 $\pm$ 0.17	31 $\pm$ 25	0.57 $\pm$ 0.11	34 $\pm$ 10
	ADE20k	^∗0.57 $\pm$ 0.16^∗	^∗27 $\pm$ 22^∗	^∗0.60 $\pm$ 0.11^∗	^∗30 $\pm$ 13^∗