The paper introduces a novel semantic segmentation algorithm that leverages a deconvolution network. The deconvolution network, built on top of a VGG 16-layer net, consists of deconvolution and unpooling layers to identify pixel-wise class labels and predict segmentation masks. The trained network is applied to individual object proposals in an input image, and the results are combined to form the final semantic segmentation map. This approach addresses the limitations of existing methods based on fully convolutional networks (FCNs) by integrating deep deconvolution and proposal-wise prediction, effectively handling objects at multiple scales and detailed structures. The proposed method achieves outstanding performance on the PASCAL VOC 2012 dataset, outperforming other methods trained without external data through an ensemble with FCN-8s. The paper also discusses the architecture of the deconvolution network, training strategies, and experimental results, highlighting the complementary characteristics of the proposed method and FCN.The paper introduces a novel semantic segmentation algorithm that leverages a deconvolution network. The deconvolution network, built on top of a VGG 16-layer net, consists of deconvolution and unpooling layers to identify pixel-wise class labels and predict segmentation masks. The trained network is applied to individual object proposals in an input image, and the results are combined to form the final semantic segmentation map. This approach addresses the limitations of existing methods based on fully convolutional networks (FCNs) by integrating deep deconvolution and proposal-wise prediction, effectively handling objects at multiple scales and detailed structures. The proposed method achieves outstanding performance on the PASCAL VOC 2012 dataset, outperforming other methods trained without external data through an ensemble with FCN-8s. The paper also discusses the architecture of the deconvolution network, training strategies, and experimental results, highlighting the complementary characteristics of the proposed method and FCN.