. 2021 Mar 4;9:40496–40510. doi: 10.1109/ACCESS.2021.3063716

TABLE 3. Result on the polyp Detection and Localisation Task on the Kvasir-SEG Dataset. Two Best Scores are Highlighted in Bold.

Method	Backbone	AP	IoU	AP₂₅	AP₅₀	AP₇₅	FPS
EfficientDet-D0 [52]	EfficientNet-b0, biFPN	0.4756	0.4322	0.6846	0.5047	0.2280	35.00
Faster R-CNN [50]	ResNet50	0.7866	0.5621	0.8947	0.8418	0.5660	8.0
RetinaNet [49]	ResNet50	0.8697	0.7313	0.9395	0.9095	0.6967	16.2
RetinaNet [49]	ResNet101	0.8745	0.7579	0.9483	0.9095	0.7132	16.8
YOLOv3+spp [55]	Darknet53	0.8105	0.8248	0.8856	0.8532	0.7586	45.01
YOLOv4 [56]	Darknet53, CSP	0.8513	0.8025	0.9123	0.8234	0.7594	48.00
ColonSegNet (Proposed)	–	0.8000	0.8100	0.9000	0.8166	0.6706	180.00