{"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/nddr-cnn-layer-wise-feature-fusing-in-multi","title":"NDDR-CNN: Layerwise Feature Fusing in Multi-Task CNNs by Neural Discriminative Dimensionality Reduction","arxiv_id":"1801.08297","date":"2018-01-25","proceeding":"CVPR 2019 6","authors":["Yuan Gao","Jiayi Ma","Mingbo Zhao","Wei Liu","Alan L. Yuille"],"abstract":"In this paper, we propose a novel Convolutional Neural Network (CNN)\nstructure for general-purpose multi-task learning (MTL), which enables\nautomatic feature fusing at every layer from different tasks. This is in\ncontrast with the most widely used MTL CNN structures which empirically or\nheuristically share features on some specific layers (e.g., share all the\nfeatures except the last convolutional layer). The proposed layerwise feature\nfusing scheme is formulated by combining existing CNN components in a novel\nway, with clear mathematical interpretability as discriminative dimensionality\nreduction, which is referred to as Neural Discriminative Dimensionality\nReduction (NDDR). Specifically, we first concatenate features with the same\nspatial resolution from different tasks according to their channel dimension.\nThen, we show that the discriminative dimensionality reduction can be fulfilled\nby 1x1 Convolution, Batch Normalization, and Weight Decay in one CNN. The use\nof existing CNN components ensures the end-to-end training and the\nextensibility of the proposed NDDR layer to various state-of-the-art CNN\narchitectures in a \"plug-and-play\" manner. The detailed ablation analysis shows\nthat the proposed NDDR layer is easy to train and also robust to different\nhyperparameters. Experiments on different task sets with various base network\narchitectures demonstrate the promising performance and desirable\ngeneralizability of our proposed method. The code of our paper is available at\nhttps://github.com/ethanygao/NDDR-CNN.","url_abs":"http://arxiv.org/abs/1801.08297v4","url_pdf":"http://arxiv.org/pdf/1801.08297v4.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":"nddr-cnn-layer-wise-feature-fusing-in-multi","repo_url":"https://github.com/ethanygao/NDDR-CNN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"multi-task-learning","task_name":"Multi-Task Learning"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"interpretability","method_name":"Interpretability"},{"method_slug":"weight-decay","method_name":"Weight Decay"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/semantic-segmentation-on-nyu-depth-v2","task":"Semantic Segmentation","dataset":"NYU Depth v2","model":"NDDR-CNN","rank_in_archive_order":101,"of":121,"metrics":{"Mean IoU":"43.3%"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1801.08297","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}