{"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/parallel-multiscale-autoregressive-density","title":"Parallel Multiscale Autoregressive Density Estimation","arxiv_id":"1703.03664","date":"2017-03-10","proceeding":"ICML 2017 8","authors":["Scott Reed","Aäron van den Oord","Nal Kalchbrenner","Sergio Gómez Colmenarejo","Ziyu Wang","Dan Belov","Nando de Freitas"],"abstract":"PixelCNN achieves state-of-the-art results in density estimation for natural\nimages. Although training is fast, inference is costly, requiring one network\nevaluation per pixel; O(N) for N pixels. This can be sped up by caching\nactivations, but still involves generating each pixel sequentially. In this\nwork, we propose a parallelized PixelCNN that allows more efficient inference\nby modeling certain pixel groups as conditionally independent. Our new PixelCNN\nmodel achieves competitive density estimation and orders of magnitude speedup -\nO(log N) sampling instead of O(N) - enabling the practical generation of\n512x512 images. We evaluate the model on class-conditional image generation,\ntext-to-image synthesis, and action-conditional video generation, showing that\nour model achieves the best results among non-pixel-autoregressive density\nmodels that allow efficient sampling.","url_abs":"http://arxiv.org/abs/1703.03664v1","url_pdf":"http://arxiv.org/pdf/1703.03664v1.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":[],"tasks":[{"task_slug":"conditional-image-generation","task_name":"Conditional Image Generation"},{"task_slug":"density-estimation","task_name":"Density Estimation"},{"task_slug":"image-compression","task_name":"Image Compression"},{"task_slug":"image-generation","task_name":"Image Generation"},{"task_slug":"video-generation","task_name":"Video Generation"}],"methods":[{"method_slug":"pixelcnn","method_name":"PixelCNN"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-compression-on-imagenet32","task":"Image Compression","dataset":"ImageNet32","model":"MS-PixelCNN","rank_in_archive_order":2,"of":5,"metrics":{"bpsp":"3.95"},"uses_additional_data":false},{"leaderboard":"/sota/image-generation-on-imagenet-64x64","task":"Image Generation","dataset":"ImageNet 64x64","model":"Parallel Multiscale","rank_in_archive_order":54,"of":65,"metrics":{"Bits per dim":"3.7"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1703.03664","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}