{"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/micronnet-a-highly-compact-deep-convolutional","title":"MicronNet: A Highly Compact Deep Convolutional Neural Network Architecture for Real-time Embedded Traffic Sign Classification","arxiv_id":"1804.00497","date":"2018-03-28","proceeding":null,"authors":["Alexander Wong","Mohammad Javad Shafiee","Michael St. Jules"],"abstract":"Traffic sign recognition is a very important computer vision task for a\nnumber of real-world applications such as intelligent transportation\nsurveillance and analysis. While deep neural networks have been demonstrated in\nrecent years to provide state-of-the-art performance traffic sign recognition,\na key challenge for enabling the widespread deployment of deep neural networks\nfor embedded traffic sign recognition is the high computational and memory\nrequirements of such networks. As a consequence, there are significant benefits\nin investigating compact deep neural network architectures for traffic sign\nrecognition that are better suited for embedded devices. In this paper, we\nintroduce MicronNet, a highly compact deep convolutional neural network for\nreal-time embedded traffic sign recognition designed based on macroarchitecture\ndesign principles (e.g., spectral macroarchitecture augmentation, parameter\nprecision optimization, etc.) as well as numerical microarchitecture\noptimization strategies. The resulting overall architecture of MicronNet is\nthus designed with as few parameters and computations as possible while\nmaintaining recognition performance, leading to optimized information density\nof the proposed network. The resulting MicronNet possesses a model size of just\n~1MB and ~510,000 parameters (~27x fewer parameters than state-of-the-art)\nwhile still achieving a human performance level top-1 accuracy of 98.9% on the\nGerman traffic sign recognition benchmark. Furthermore, MicronNet requires just\n~10 million multiply-accumulate operations to perform inference, and has a\ntime-to-compute of just 32.19 ms on a Cortex-A53 high efficiency processor.\nThese experimental results show that highly compact, optimized deep neural\nnetwork architectures can be designed for real-time traffic sign recognition\nthat are well-suited for embedded scenarios.","url_abs":"http://arxiv.org/abs/1804.00497v3","url_pdf":"http://arxiv.org/pdf/1804.00497v3.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":"micronnet-a-highly-compact-deep-convolutional","repo_url":"https://github.com/ppriyank/MicronNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"traffic-sign-recognition","task_name":"Traffic Sign Recognition"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/traffic-sign-recognition-on-gtsrb","task":"Traffic Sign Recognition","dataset":"GTSRB","model":"MicronNet (fp16)","rank_in_archive_order":4,"of":5,"metrics":{"Accuracy":"98.9%"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}