{"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/dense-and-diverse-capsule-networks-making-the","title":"Dense and Diverse Capsule Networks: Making the Capsules Learn Better","arxiv_id":"1805.04001","date":"2018-05-10","proceeding":null,"authors":["Sai Samarth R. Phaye","Apoorva Sikka","Abhinav Dhall","Deepti Bathula"],"abstract":"Past few years have witnessed exponential growth of interest in deep learning\nmethodologies with rapidly improving accuracies and reduced computational\ncomplexity. In particular, architectures using Convolutional Neural Networks\n(CNNs) have produced state-of-the-art performances for image classification and\nobject recognition tasks. Recently, Capsule Networks (CapsNet) achieved\nsignificant increase in performance by addressing an inherent limitation of\nCNNs in encoding pose and deformation. Inspired by such advancement, we asked\nourselves, can we do better? We propose Dense Capsule Networks (DCNet) and\nDiverse Capsule Networks (DCNet++). The two proposed frameworks customize the\nCapsNet by replacing the standard convolutional layers with densely connected\nconvolutions. This helps in incorporating feature maps learned by different\nlayers in forming the primary capsules. DCNet, essentially adds a deeper\nconvolution network, which leads to learning of discriminative feature maps.\nAdditionally, DCNet++ uses a hierarchical architecture to learn capsules that\nrepresent spatial information in a fine-to-coarser manner, which makes it more\nefficient for learning complex data. Experiments on image classification task\nusing benchmark datasets demonstrate the efficacy of the proposed\narchitectures. DCNet achieves state-of-the-art performance (99.75%) on MNIST\ndataset with twenty fold decrease in total training iterations, over the\nconventional CapsNet. Furthermore, DCNet++ performs better than CapsNet on SVHN\ndataset (96.90%), and outperforms the ensemble of seven CapsNet models on\nCIFAR-10 by 0.31% with seven fold decrease in number of parameters.","url_abs":"http://arxiv.org/abs/1805.04001v1","url_pdf":"http://arxiv.org/pdf/1805.04001v1.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":"dense-and-diverse-capsule-networks-making-the","repo_url":"https://github.com/ssrp/Multi-level-DCNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"GPL-3.0"}}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"object-recognition","task_name":"Object Recognition"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-classification-on-smallnorb","task":"Image Classification","dataset":"smallNORB","model":"DCNet","rank_in_archive_order":7,"of":7,"metrics":{"Classification Error":"5.57"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1805.04001","atlas_url":"https://app.syntology.ai/?focus=1805.04001","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}