{"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/bayesian-prediction-of-future-street-scenes-1","title":"Bayesian Prediction of Future Street Scenes using Synthetic Likelihoods","arxiv_id":"1810.00746","date":"2018-10-01","proceeding":"ICLR 2019 5","authors":["Apratim Bhattacharyya","Mario Fritz","Bernt Schiele"],"abstract":"For autonomous agents to successfully operate in the real world, the ability\nto anticipate future scene states is a key competence. In real-world scenarios,\nfuture states become increasingly uncertain and multi-modal, particularly on\nlong time horizons. Dropout based Bayesian inference provides a computationally\ntractable, theoretically well grounded approach to learn likely\nhypotheses/models to deal with uncertain futures and make predictions that\ncorrespond well to observations -- are well calibrated. However, it turns out\nthat such approaches fall short to capture complex real-world scenes, even\nfalling behind in accuracy when compared to the plain deterministic approaches.\nThis is because the used log-likelihood estimate discourages diversity. In this\nwork, we propose a novel Bayesian formulation for anticipating future scene\nstates which leverages synthetic likelihoods that encourage the learning of\ndiverse models to accurately capture the multi-modal nature of future scene\nstates. We show that our approach achieves accurate state-of-the-art\npredictions and calibrated probabilities through extensive experiments for\nscene anticipation on Cityscapes dataset. Moreover, we show that our approach\ngeneralizes across diverse tasks such as digit generation and precipitation\nforecasting.","url_abs":"http://arxiv.org/abs/1810.00746v3","url_pdf":"http://arxiv.org/pdf/1810.00746v3.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":"bayesian-prediction-of-future-street-scenes-1","repo_url":"https://github.com/apratimbhattacharyya18/seg_pred","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"tf","reach":{"status":"ok"}}],"tasks":[{"task_slug":"bayesian-inference","task_name":"Bayesian Inference"},{"task_slug":"diversity","task_name":"Diversity"},{"task_slug":"precipitation-forecasting","task_name":"Precipitation Forecasting"}],"methods":[{"method_slug":"dropout","method_name":"Dropout"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1810.00746","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}