Table 8.

Binary segmentation results (mIoU) for models pre-trained on EndoVis17 and fine-tuned for each of the target datasets (Kvasir and Endomapper).

Models	Datasets			Computational cost
	EndoVis17	Kvasir-Inst.	Endomapper	Params (M)+	Time (ms)++
U-Net39	75.44	85.78	55.63	7.85	54
TernausNet40	83.60	N.A.	N.A.	36.92	119
LinkNet41	82.36	87.75	60.54	21.79	34
MF-TAPNet42	87.56	86.81	66.87	37.73	155
MiniNet v243	87.16	85.13	66.65	0.52	26

N.A.: Not available due to computational resource limitations.

⁺Params: Memory required by the model (M = millions of parameters).

⁺⁺Time: Average inference time for 1 image on GPU RTX2080.