Progressive CNN-transformer semantic compensation network for polyp segmentation

Table 1. Introduction to the dataset
View table
View in Article
Table 1. Introduction to the dataset
数据集年份原始分辨率图像总数
CVC-ClinicDB 2015 384×288 612
Kvasir-SEG 2020 332×487~1 920×1 072 1 000
CVC-300 2017 574×500 60
CVC-ColonDB 2016 574×500 380

Table 2. Performance comparison of different methods on CVC-ClinicDB and Kvasir-SEG datasets

View table

View in Article

Table 2. Performance comparison of different methods on CVC-ClinicDB and Kvasir-SEG datasets

Methods	CVC-ClinicDB						Kvasir-SEG
Methods	$m D i c e$	$m I o U p$	$M A E$	$S_{α}$	$F_{β}^{ω}$	$E_{ϕ}^{}$	$m D i c e$	$m I o U p$	$M A E$	$S_{α}$	$F_{β}^{ω}$	$E_{ϕ}^{}$
U-Net^［32］	0.901	0.849	0.011	0.921	0.895	0.971	0.844	0.772	0.042	0.873	0.819	0.901
U-Net++^［33］	0.911	0.856	0.013	0.929	0.904	0.965	0.840	0.769	0.044	0.872	0.813	0.901
CE-Net^［34］	0.914	0.867	0.015	0.933	0.911	0.965	0.894	0.835	0.034	0.903	0.877	0.908
PraNet^［5］	0.930	0.887	0.008	0.943	0.919	0.964	0.894	0.842	0.033	0.901	0.881	0.938
UACANet^［15］	0.937	0.889	0.008	0.941	0.929	0.982	0.906	0.855	0.026	0.914	0.897	0.948
DCRNet^［8］	0.935	0.895	0.008	0.943	0.927	0.985	0.882	0.831	0.033	0.901	0.870	0.917
CFA-Net^［9］	0.865	0.810	0.017	0.902	0.863	0.959	0.893	0.837	0.026	0.906	0.878	0.923
META-Unet^［35］	0.912	0.861	0.009	0.929	0.907	0.975	0.898	0.843	0.029	0.911	0.884	0.926
SSFormer^［10］	0.893	0.840	0.017	0.920	0.884	0.950	0.917	0.866	0.022	0.924	0.910	0.940
Colonformer^［37］	0.930	0.891	0.007	0.946	0.920	0.966	0.922	0.877	0.021	0.929	0.908	0.948
MCE-Net^［16］	0.925	0.877	0.008	0.939	0.917	0.973	0.912	0.865	0.023	0.920	0.903	0.935
CGMA-Net^［36］	0.939	0.903	0.007	0.951	0.933	0.984	0.918	0.874	0.024	0.927	0.901	0.941
FCBFormer^［14］	0.897	0.844	0.014	0.921	0.886	0.955	0.913	0.860	0.027	0.915	0.899	0.942
TransFuse^［13］	0.880	0.821	0.017	0.913	0.869	0.952	0.872	0.809	0.035	0.890	0.855	0.923
DFETC-Net^［21］	0.893	0.846	0.010	0.924	0.889	0.957	0.899	0.839	0.028	0.906	0.883	0.936
MC-DC^［38］	0.930	0.881	0.011	0.937	0.927	0.981	0.907	0.853	0.027	0.915	0.898	0.937
TGDAUNet^［39］	0.911	0.861	0.009	0.928	0.910	0.966	0.898	0.843	0.031	0.907	0.887	0.934
PCTSC	0.942	0.899	0.006	0.946	0.937	0.985	0.929	0.882	0.021	0.929	0.919	0.952

Table 3. Performance comparison of different methods on CVC-300 and CVC-ColonDB datasets

View table

View in Article

Table 3. Performance comparison of different methods on CVC-300 and CVC-ColonDB datasets

Methods	CVC-300						CVC-ColonDB
Methods	$m D i c e$	$m I o U p$	$M A E$	$S_{α}$	$F_{β}^{ω}$	$E_{ϕ}^{}$	$m D i c e$	$m I o U p$	$M A E$	$S_{α}$	$F_{β}^{ω}$	$E_{ϕ}^{}$
U-Net^［32］	0.670	0.588	0.019	0.802	0.636	0.880	0.588	0.507	0.051	0.741	0.563	0.813
U-Net++^［33］	0.709	0.627	0.017	0.827	0.666	0.863	0.591	0.511	0.049	0.747	0.566	0.807
CE-Net^［34］	0.893	0.825	0.008	0.930	0.870	0.960	0.732	0.655	0.047	0.824	0.710	0.858
PraNet^［5］	0.891	0.828	0.008	0.934	0.869	0.952	0.713	0.641	0.042	0.816	0.693	0.845
UACANet^［15］	0.879	0.802	0.009	0.918	0.844	0.950	0.734	0.662	0.037	0.828	0.721	0.885
DCRNet^［8］	0.871	0.806	0.007	0.918	0.841	0.960	0.650	0.584	0.050	0.782	0.639	0.844
CFA-Net^［9］	0.885	0.818	0.007	0.928	0.861	0.961	0.667	0.591	0.048	0.789	0.656	0.868
META-Unet^［35］	0.892	0.819	0.008	0.925	0.868	0.965	0.688	0.614	0.044	0.800	0.672	0.853
SSFormer^［10］	0.887	0.815	0.009	0.927	0.861	0.953	0.779	0.696	0.034	0.845	0.753	0.883
Colonformer^［37］	0.875	0.804	0.008	0.921	0.848	0.949	0.807	0.731	0.031	0.867	0.784	0.901
MCE-Net^［16］	0.890	0.824	0.010	0.928	0.868	0.955	0.791	0.711	0.032	0.851	0.775	0.903
CGMA-Net^［36］	0.891	0.829	0.008	0.935	0.865	0.961	0.792	0.721	0.033	0.863	0.772	0.896
FCBFormer^［14］	0.878	0.806	0.010	0.919	0.847	0.948	0.787	0.706	0.034	0.848	0.756	0.884
TransFuse^［13］	0.850	0.777	0.012	0.913	0.817	0.921	0.734	0.646	0.036	0.825	0.707	0.869
DFETC-Net^［21］	0.845	0.766	0.014	0.907	0.805	0.921	0.653	0.572	0.051	0.782	0.636	0.851
MC-DC^［38］	0.888	0.821	0.008	0.933	0.862	0.947	0.778	0.691	0.030	0.846	0.754	0.894
TGDAUNet^［39］	0.894	0.828	0.008	0.933	0.870	0.957	0.756	0.681	0.035	0.842	0.741	0.883
PCTSC	0.904	0.838	0.006	0.936	0.882	0.969	0.803	0.721	0.029	0.860	0.781	0.910

Table 4. Comparison of computational efficiency of different methods

View table

View in Article

Table 4. Comparison of computational efficiency of different methods

Methods	FLOPs/G	Params/M
U-Net^［32］	152.154	39.396
U-Net++^［33］	377.449	47.176
CE-Net^［34］	16.828	29.003
PraNet^［5］	13.149	30.498
UACANet^［15］	59.646	67.109
DCRNet^［8］	17.268	28.734
CFA-Net^［9］	55.360	25.239
META-Unet^［35］	9.716	21.696
SSFormer^［10］	22.746	46.399
Colonformer^［37］	22.982	52.945
MCE-Net^［16］	39.212	58.459
CGMA-Net^［36］	33.121	30.962
FCBFormer^［14］	73.295	52.943
TransFuse^［13］	21.788	26.174
DFETC-Net^［21］	66.013	46.092
MC-DC^［38］	126.950	100.979
TGDAUNet^［39］	105.145	261.502
PCTSC	16.264	34.586

Table 5. Results of ablation experiments for different modules

View table

View in Article

Table 5. Results of ablation experiments for different modules

Methods	CVC-ClinicDB		Kvasir-SEG		CVC-300		CVC-ColonDB
Methods	mDice	mIoUp	mDice	mIoUp	mDice	mIoUp	mDice	mIoUp
Baseline	0.933	0.886	0.915	0.861	0.883	0.814	0.793	0.701
Baseline+SLFIC	0.939	0.893	0.922	0.873	0.902	0.837	0.815	0.734
Baseline+QBSC	0.934	0.890	0.927	0.879	0.903	0.837	0.798	0.721
PCTSC	0.942	0.899	0.929	0.882	0.904	0.838	0.803	0.721

Tools

Get Citation

Copy Citation Text

Daxiang LI, Denghui LI, Ying LIU, Yao TANG. Progressive CNN-transformer semantic compensation network for polyp segmentation[J]. Optics and Precision Engineering, 2024, 32(16): 2523

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites