ELMGAN: A GAN-based efficient lightweight multi-scale-feature-fusion multi-task model
作者：
Highlights：
• We proposed a non-Point based counting method based on the number of cells and the global segmentation.
• We proposed ELMGAN to segment and count cells, and proposed the NH Loss to train our network.
• Our new upsampling method: FBU is twice as fast as the traditional interpolation methods.
• Our generator: LFMMG is only one tenth the size of traditional Gan, but ten times faster.
• Our ELMGAN overcomes the limitation of Point-based counting methods, and is better than SOTA.
摘要
•We proposed a non-Point based counting method based on the number of cells and the global segmentation.•We proposed ELMGAN to segment and count cells, and proposed the NH Loss to train our network.•Our new upsampling method: FBU is twice as fast as the traditional interpolation methods.•Our generator: LFMMG is only one tenth the size of traditional Gan, but ten times faster.•Our ELMGAN overcomes the limitation of Point-based counting methods, and is better than SOTA.
论文关键词：Convolutional neural network,Generative adversarial networks,Cell segmentation,Cell counting
论文评审过程：Received 22 April 2022, Revised 11 July 2022, Accepted 11 July 2022, Available online 16 July 2022, Version of Record 27 July 2022.
论文官网地址：https://doi.org/10.1016/j.knosys.2022.109434
