Automated cervical cell segmentation using deep ensemble learning

BMC Medical Imaging

Table 1 Characteristics of candidate base models

Model	Parameters	Encoder	Decoder
Unet_resnet34	24,436,369	33 conv layers	10 conv layers
Unet_densenet121	13,607,633	117 conv layers	10 conv layers
UnetPlusPlus_resnet34	26,078,609	33 conv layers	10 conv layers
UnetPlusPlus_densenet121	30,072,273	117 conv layers	10 conv layers
DeepLabV3_resnet34	26,007,105	33 conv layers	1 conv layer + ASPP
DeepLabV3_resnet50	39,633,729	49 conv layers	1 conv layer + ASPP
DeepLabV3Plus_resnet34	22,437,457	33 conv layers	2conv layers + ASPP
DeepLabV3Plus_resnet50	26,677,585	49 conv layers	2conv layers + ASPP
Segformer	7,717,473	/	/
Transunet	67,875,963	/	/

Conv layer and ASPP stand for convolutional layer and atrous spatial pyramid pooling layer, respectively. Segformer and Transunet are transformer-based models, their encoder and decoder structures are not listed

ISSN: 1471-2342