{"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/sdcnet-video-prediction-using-spatially","title":"SDCNet: Video Prediction Using Spatially-Displaced Convolution","arxiv_id":"1811.00684","date":"2018-11-02","proceeding":null,"authors":["Fitsum A. Reda","Guilin Liu","Kevin J. Shih","Robert Kirby","Jon Barker","David Tarjan","Andrew Tao","Bryan Catanzaro"],"abstract":"We present an approach for high-resolution video frame prediction by\nconditioning on both past frames and past optical flows. Previous approaches\nrely on resampling past frames, guided by a learned future optical flow, or on\ndirect generation of pixels. Resampling based on flow is insufficient because\nit cannot deal with disocclusions. Generative models currently lead to blurry\nresults. Recent approaches synthesis a pixel by convolving input patches with a\npredicted kernel. However, their memory requirement increases with kernel size.\nHere, we spatially-displaced convolution (SDC) module for video frame\nprediction. We learn a motion vector and a kernel for each pixel and synthesize\na pixel by applying the kernel at a displaced location in the source image,\ndefined by the predicted motion vector. Our approach inherits the merits of\nboth vector-based and kernel-based approaches, while ameliorating their\nrespective disadvantages. We train our model on 428K unlabelled 1080p video\ngame frames. Our approach produces state-of-the-art results, achieving an SSIM\nscore of 0.904 on high-definition YouTube-8M videos, 0.918 on Caltech\nPedestrian videos. Our model handles large motion effectively and synthesizes\ncrisp frames with consistent motion.","url_abs":"http://arxiv.org/abs/1811.00684v1","url_pdf":"http://arxiv.org/pdf/1811.00684v1.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":"sdcnet-video-prediction-using-spatially","repo_url":"https://github.com/NVIDIA/semantic-segmentation","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"BSD-3-Clause"}},{"paper_slug":"sdcnet-video-prediction-using-spatially","repo_url":"https://github.com/yangyucheng000/papercode-2/tree/main/SDCF-mindspore-main","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[{"task_slug":"optical-flow-estimation","task_name":"Optical Flow Estimation"},{"task_slug":"prediction","task_name":"Prediction"},{"task_slug":"ssim","task_name":"SSIM"},{"task_slug":"video-prediction","task_name":"Video Prediction"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1811.00684","atlas_url":"https://app.syntology.ai/?focus=1811.00684","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}