(57) A method for improving image segmentation by using a learning device is disclosed. The method includes steps of: (a) if a training image is obtained, acquiring (2- K)_th to (2-1)_th feature maps through an encoding layer and a decoding layer, and acquiring 1_st to H_th losses from the 1_st to the H_th loss layers respectively corresponding to H feature maps, obtained from the H filters, among the (2-K)_th to the (2-1)_th feature maps; and (b) upon performing a backpropagation process, performing processes of allowing the (2-M)_th filter to apply a convolution operation to (M-1)₂-th adjusted feature map relayed from the (2-(M-1))_th filter to obtain M₁-th temporary feature map; relaying, to the (2-(M+1))_th filter, M₂-th adjusted feature map obtained by computing the M_th loss with the M₁-th temporary feature map; and adjusting at least part of parameters of the (1-1)_th to the (1-K)_th filters and the (2-K)_th to the (2-1)_th filters.