{"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/tramnet-transition-matrix-network-for","title":"TraMNet - Transition Matrix Network for Efficient Action Tube Proposals","arxiv_id":"1808.00297","date":"2018-08-01","proceeding":null,"authors":["Gurkirt Singh","Suman Saha","Fabio Cuzzolin"],"abstract":"Current state-of-the-art methods solve spatiotemporal action localisation by\nextending 2D anchors to 3D-cuboid proposals on stacks of frames, to generate\nsets of temporally connected bounding boxes called \\textit{action micro-tubes}.\nHowever, they fail to consider that the underlying anchor proposal hypotheses\nshould also move (transition) from frame to frame, as the actor or the camera\ndoes. Assuming we evaluate $n$ 2D anchors in each frame, then the number of\npossible transitions from each 2D anchor to the next, for a sequence of $f$\nconsecutive frames, is in the order of $O(n^f)$, expensive even for small\nvalues of $f$. To avoid this problem, we introduce a Transition-Matrix-based\nNetwork (TraMNet) which relies on computing transition probabilities between\nanchor proposals while maximising their overlap with ground truth bounding\nboxes across frames, and enforcing sparsity via a transition threshold. As the\nresulting transition matrix is sparse and stochastic, this reduces the proposal\nhypothesis search space from $O(n^f)$ to the cardinality of the thresholded\nmatrix. At training time, transitions are specific to cell locations of the\nfeature maps, so that a sparse (efficient) transition matrix is used to train\nthe network. At test time, a denser transition matrix can be obtained either by\ndecreasing the threshold or by adding to it all the relative transitions\noriginating from any cell location, allowing the network to handle transitions\nin the test data that might not have been present in the training data, and\nmaking detection translation-invariant. Finally, we show that our network can\nhandle sparse annotations such as those available in the DALY dataset. We\nreport extensive experiments on the DALY, UCF101-24 and Transformed-UCF101-24\ndatasets to support our claims.","url_abs":"http://arxiv.org/abs/1808.00297v1","url_pdf":"http://arxiv.org/pdf/1808.00297v1.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":"tramnet-transition-matrix-network-for","repo_url":"https://github.com/gurkirt/AMTNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}