{"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/fcn-rlstm-deep-spatio-temporal-neural","title":"FCN-rLSTM: Deep Spatio-Temporal Neural Networks for Vehicle Counting in City Cameras","arxiv_id":"1707.09476","date":"2017-07-29","proceeding":"ICCV 2017 10","authors":["Shanghang Zhang","Guanhang Wu","João P. Costeira","José M. F. Moura"],"abstract":"In this paper, we develop deep spatio-temporal neural networks to\nsequentially count vehicles from low quality videos captured by city cameras\n(citycams). Citycam videos have low resolution, low frame rate, high occlusion\nand large perspective, making most existing methods lose their efficacy. To\novercome limitations of existing methods and incorporate the temporal\ninformation of traffic video, we design a novel FCN-rLSTM network to jointly\nestimate vehicle density and vehicle count by connecting fully convolutional\nneural networks (FCN) with long short term memory networks (LSTM) in a residual\nlearning fashion. Such design leverages the strengths of FCN for pixel-level\nprediction and the strengths of LSTM for learning complex temporal dynamics.\nThe residual learning connection reformulates the vehicle count regression as\nlearning residual functions with reference to the sum of densities in each\nframe, which significantly accelerates the training of networks. To preserve\nfeature map resolution, we propose a Hyper-Atrous combination to integrate\natrous convolution in FCN and combine feature maps of different convolution\nlayers. FCN-rLSTM enables refined feature representation and a novel end-to-end\ntrainable mapping from pixels to vehicle count. We extensively evaluated the\nproposed method on different counting tasks with three datasets, with\nexperimental results demonstrating their effectiveness and robustness. In\nparticular, FCN-rLSTM reduces the mean absolute error (MAE) from 5.31 to 4.21\non TRANCOS, and reduces the MAE from 2.74 to 1.53 on WebCamT. Training process\nis accelerated by 5 times on average.","url_abs":"http://arxiv.org/abs/1707.09476v2","url_pdf":"http://arxiv.org/pdf/1707.09476v2.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":"fcn-rlstm-deep-spatio-temporal-neural","repo_url":"https://github.com/dpernes/FCN-rLSTM","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[],"methods":[{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"fcn","method_name":"FCN"},{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1707.09476","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}