{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/pixel-deconvolutional-networks","title":"Pixel Deconvolutional Networks","arxiv_id":"1705.06820","date":"2017-05-18","proceeding":"ICLR 2018 1","authors":["Hongyang Gao","Hao Yuan","Zhengyang Wang","Shuiwang Ji"],"abstract":"Deconvolutional layers have been widely used in a variety of deep models for\nup-sampling, including encoder-decoder networks for semantic segmentation and\ndeep generative models for unsupervised learning. One of the key limitations of\ndeconvolutional operations is that they result in the so-called checkerboard\nproblem. This is caused by the fact that no direct relationship exists among\nadjacent pixels on the output feature map. To address this problem, we propose\nthe pixel deconvolutional layer (PixelDCL) to establish direct relationships\namong adjacent pixels on the up-sampled feature map. Our method is based on a\nfresh interpretation of the regular deconvolution operation. The resulting\nPixelDCL can be used to replace any deconvolutional layer in a plug-and-play\nmanner without compromising the fully trainable capabilities of original\nmodels. The proposed PixelDCL may result in slight decrease in efficiency, but\nthis can be overcome by an implementation trick. Experimental results on\nsemantic segmentation demonstrate that PixelDCL can consider spatial features\nsuch as edges and shapes and yields more accurate segmentation outputs than\ndeconvolutional layers. When used in image generation tasks, our PixelDCL can\nlargely overcome the checkerboard problem suffered by regular deconvolution\noperations.","url_abs":"http://arxiv.org/abs/1705.06820v4","url_pdf":"http://arxiv.org/pdf/1705.06820v4.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"pixel-deconvolutional-networks","repo_url":"https://github.com/divelab/PixelDCN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"pixel-deconvolutional-networks","repo_url":"https://github.com/HongyangGao/PixelTCN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"pixel-deconvolutional-networks","repo_url":"https://github.com/fourmi1995/IronsegExperiment-PixelDCL","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"pixel-deconvolutional-networks","repo_url":"https://github.com/kaustubh1955/pixelDeconvolution","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"pixel-deconvolutional-networks","repo_url":"https://github.com/zhengyang-wang/Unet_3D","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"image-generation","task_name":"Image Generation"},{"task_slug":"segmentation","task_name":"Segmentation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1705.06820","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}