{"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/robust-visual-tracking-via-hierarchical","title":"Robust Visual Tracking via Hierarchical Convolutional Features","arxiv_id":"1707.03816","date":"2017-07-12","proceeding":null,"authors":["Chao Ma","Jia-Bin Huang","Xiaokang Yang","Ming-Hsuan Yang"],"abstract":"In this paper, we propose to exploit the rich hierarchical features of deep\nconvolutional neural networks to improve the accuracy and robustness of visual\ntracking. Deep neural networks trained on object recognition datasets consist\nof multiple convolutional layers. These layers encode target appearance with\ndifferent levels of abstraction. For example, the outputs of the last\nconvolutional layers encode the semantic information of targets and such\nrepresentations are invariant to significant appearance variations. However,\ntheir spatial resolutions are too coarse to precisely localize the target. In\ncontrast, features from earlier convolutional layers provide more precise\nlocalization but are less invariant to appearance changes. We interpret the\nhierarchical features of convolutional layers as a nonlinear counterpart of an\nimage pyramid representation and explicitly exploit these multiple levels of\nabstraction to represent target objects. Specifically, we learn adaptive\ncorrelation filters on the outputs from each convolutional layer to encode the\ntarget appearance. We infer the maximum response of each layer to locate\ntargets in a coarse-to-fine manner. To further handle the issues with scale\nestimation and re-detecting target objects from tracking failures caused by\nheavy occlusion or out-of-the-view movement, we conservatively learn another\ncorrelation filter, that maintains a long-term memory of target appearance, as\na discriminative classifier. We apply the classifier to two types of object\nproposals: (1) proposals with a small step size and tightly around the\nestimated location for scale estimation; and (2) proposals with large step size\nand across the whole image for target re-detection. Extensive experimental\nresults on large-scale benchmark datasets show that the proposed algorithm\nperforms favorably against state-of-the-art tracking methods.","url_abs":"http://arxiv.org/abs/1707.03816v2","url_pdf":"http://arxiv.org/pdf/1707.03816v2.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":"robust-visual-tracking-via-hierarchical","repo_url":"https://github.com/chaoma99/HCFTstar","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"object-recognition","task_name":"Object Recognition"},{"task_slug":"visual-tracking","task_name":"Visual Tracking"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}