{"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/smoothed-dilated-convolutions-for-improved","title":"Smoothed Dilated Convolutions for Improved Dense Prediction","arxiv_id":"1808.08931","date":"2018-08-27","proceeding":null,"authors":["Zhengyang Wang","Shuiwang Ji"],"abstract":"Dilated convolutions, also known as atrous convolutions, have been widely\nexplored in deep convolutional neural networks (DCNNs) for various dense\nprediction tasks. However, dilated convolutions suffer from the gridding\nartifacts, which hampers the performance. In this work, we propose two simple\nyet effective degridding methods by studying a decomposition of dilated\nconvolutions. Unlike existing models, which explore solutions by focusing on a\nblock of cascaded dilated convolutional layers, our methods address the\ngridding artifacts by smoothing the dilated convolution itself. In addition, we\npoint out that the two degridding approaches are intrinsically related and\ndefine separable and shared (SS) operations, which generalize the proposed\nmethods. We further explore SS operations in view of operations on graphs and\npropose the SS output layer, which is able to smooth the entire DCNNs by only\nreplacing the output layer. We evaluate our degridding methods and the SS\noutput layer thoroughly, and visualize the smoothing effect through effective\nreceptive field analysis. Results show that our methods degridding yield\nconsistent improvements on the performance of dense prediction tasks, while\nadding negligible amounts of extra training parameters. And the SS output layer\nimproves the performance significantly and is very efficient in terms of number\nof training parameters.","url_abs":"http://arxiv.org/abs/1808.08931v2","url_pdf":"http://arxiv.org/pdf/1808.08931v2.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":"smoothed-dilated-convolutions-for-improved","repo_url":"https://github.com/divelab/dilated","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"audio-generation","task_name":"Audio Generation"},{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"prediction","task_name":"Prediction"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dilated-convolution","method_name":"Dilated Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.08931","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}