{"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/understanding-the-impact-of-label-granularity","title":"Understanding the Impact of Label Granularity on CNN-based Image Classification","arxiv_id":"1901.07012","date":"2019-01-21","proceeding":null,"authors":["Zhuo Chen","Ruizhou Ding","Ting-Wu Chin","Diana Marculescu"],"abstract":"In recent years, supervised learning using Convolutional Neural Networks\n(CNNs) has achieved great success in image classification tasks, and large\nscale labeled datasets have contributed significantly to this achievement.\nHowever, the definition of a label is often application dependent. For example,\nan image of a cat can be labeled as \"cat\" or perhaps more specifically \"Persian\ncat.\" We refer to this as label granularity. In this paper, we conduct\nextensive experiments using various datasets to demonstrate and analyze how and\nwhy training based on fine-grain labeling, such as \"Persian cat\" can improve\nCNN accuracy on classifying coarse-grain classes, in this case \"cat.\" The\nexperimental results show that training CNNs with fine-grain labels improves\nboth network's optimization and generalization capabilities, as intuitively it\nencourages the network to learn more features, and hence increases\nclassification accuracy on coarse-grain classes under all datasets considered.\nMoreover, fine-grain labels enhance data efficiency in CNN training. For\nexample, a CNN trained with fine-grain labels and only 40% of the total\ntraining data can achieve higher accuracy than a CNN trained with the full\ntraining dataset and coarse-grain labels. These results point to two possible\napplications of this work: (i) with sufficient human resources, one can improve\nCNN performance by re-labeling the dataset with fine-grain labels, and (ii)\nwith limited human resources, to improve CNN performance, rather than\ncollecting more training data, one may instead use fine-grain labels for the\ndataset. We further propose a metric called Average Confusion Ratio to\ncharacterize the effectiveness of fine-grain labeling, and show its use through\nextensive experimentation. Code is available at\nhttps://github.com/cmu-enyac/Label-Granularity.","url_abs":"http://arxiv.org/abs/1901.07012v1","url_pdf":"http://arxiv.org/pdf/1901.07012v1.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":"understanding-the-impact-of-label-granularity","repo_url":"https://github.com/cmu-enyac/Label-Granularity","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1901.07012","atlas_url":"https://app.syntology.ai/?focus=1901.07012","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}