{"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/efficient-convolutional-neural-networks-for","title":"Efficient Convolutional Neural Networks for Pixelwise Classification on Heterogeneous Hardware Systems","arxiv_id":"1509.03371","date":"2015-09-11","proceeding":null,"authors":["Fabian Tschopp"],"abstract":"This work presents and analyzes three convolutional neural network (CNN)\nmodels for efficient pixelwise classification of images. When using\nconvolutional neural networks to classify single pixels in patches of a whole\nimage, a lot of redundant computations are carried out when using sliding\nwindow networks. This set of new architectures solve this issue by either\nremoving redundant computations or using fully convolutional architectures that\ninherently predict many pixels at once.\n  The implementations of the three models are accessible through a new utility\non top of the Caffe library. The utility provides support for a wide range of\nimage input and output formats, pre-processing parameters and methods to\nequalize the label histogram during training. The Caffe library has been\nextended by new layers and a new backend for availability on a wider range of\nhardware such as CPUs and GPUs through OpenCL.\n  On AMD GPUs, speedups of $54\\times$ (SK-Net), $437\\times$ (U-Net) and\n$320\\times$ (USK-Net) have been observed, taking the SK equivalent SW (sliding\nwindow) network as the baseline. The label throughput is up to one megapixel\nper second.\n  The analyzed neural networks have distinctive characteristics that apply\nduring training or processing, and not every data set is suitable to every\narchitecture. The quality of the predictions is assessed on two neural tissue\ndata sets, of which one is the ISBI 2012 challenge data set. Two different loss\nfunctions, Malis loss and Softmax loss, were used during training.\n  The whole pipeline, consisting of models, interface and modified Caffe\nlibrary, is available as Open Source software under the working title Project\nGreentea.","url_abs":"http://arxiv.org/abs/1509.03371v1","url_pdf":"http://arxiv.org/pdf/1509.03371v1.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":"efficient-convolutional-neural-networks-for","repo_url":"https://github.com/naibaf7/caffe","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"efficient-convolutional-neural-networks-for","repo_url":"https://github.com/naibaf7/caffe_neural_models","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"efficient-convolutional-neural-networks-for","repo_url":"https://github.com/naibaf7/caffe_neural_tool","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"classification","task_name":"General Classification"}],"methods":[{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}