{"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/xmem-long-term-video-object-segmentation-with","title":"XMem: Long-Term Video Object Segmentation with an Atkinson-Shiffrin Memory Model","arxiv_id":"2207.07115","date":"2022-07-14","proceeding":null,"authors":["Ho Kei Cheng","Alexander G. Schwing"],"abstract":"We present XMem, a video object segmentation architecture for long videos with unified feature memory stores inspired by the Atkinson-Shiffrin memory model. Prior work on video object segmentation typically only uses one type of feature memory. For videos longer than a minute, a single feature memory model tightly links memory consumption and accuracy. In contrast, following the Atkinson-Shiffrin model, we develop an architecture that incorporates multiple independent yet deeply-connected feature memory stores: a rapidly updated sensory memory, a high-resolution working memory, and a compact thus sustained long-term memory. Crucially, we develop a memory potentiation algorithm that routinely consolidates actively used working memory elements into the long-term memory, which avoids memory explosion and minimizes performance decay for long-term prediction. Combined with a new memory reading mechanism, XMem greatly exceeds state-of-the-art performance on long-video datasets while being on par with state-of-the-art methods (that do not work on long videos) on short-video datasets. Code is available at https://hkchengrex.github.io/XMem","url_abs":"https://arxiv.org/abs/2207.07115v2","url_pdf":"https://arxiv.org/pdf/2207.07115v2.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":"xmem-long-term-video-object-segmentation-with","repo_url":"https://github.com/hkchengrex/XMem","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"xmem-long-term-video-object-segmentation-with","repo_url":"https://github.com/tianyuan168326/videosemanticcompression-pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"2d-human-pose-estimation","task_name":"2D Human Pose Estimation"},{"task_slug":"2d-object-detection","task_name":"2D Object Detection"},{"task_slug":"3d-absolute-human-pose-estimation","task_name":"3D Absolute Human Pose Estimation"},{"task_slug":"segmentation","task_name":"Segmentation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"semi-supervised-video-object-segmentation","task_name":"Semi-Supervised Video Object Segmentation"},{"task_slug":"video-object-segmentation","task_name":"Video Object Segmentation"},{"task_slug":"video-semantic-segmentation","task_name":"Video Semantic Segmentation"}],"methods":[{"method_slug":"adam","method_name":"Adam"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-20","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS (no YouTube-VOS training)","model":"XMem","rank_in_archive_order":26,"of":26,"metrics":{"FPS":"29.6"},"uses_additional_data":false},{"leaderboard":"/sota/visual-object-tracking-on-davis-2016","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem (BL30K, MS)","rank_in_archive_order":2,"of":78,"metrics":{"F-measure (Mean)":"94.4","J&F":"93.3","Jaccard (Mean)":"92.2"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2016","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem (MS)","rank_in_archive_order":6,"of":78,"metrics":{"F-measure (Mean)":"93.5","J&F":"92.7","Jaccard (Mean)":"92.0"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2016","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem (BL30K)","rank_in_archive_order":12,"of":78,"metrics":{"F-measure (Mean)":"93.2","J&F":"92.0","Jaccard (Mean)":"90.7","Speed (FPS)":"29.6"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2016","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem","rank_in_archive_order":19,"of":78,"metrics":{"F-measure (Mean)":"92.7","J&F":"91.5","Jaccard (Mean)":"90.4","Speed (FPS)":"29.6"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2016","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem (DAVIS+YouTubeVOS only)","rank_in_archive_order":25,"of":78,"metrics":{"F-measure (Mean)":"91.9","J&F":"90.8","Jaccard (Mean)":"89.6","Speed (FPS)":"29.6"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2016","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem (DAVIS only)","rank_in_archive_order":42,"of":78,"metrics":{"F-measure (Mean)":"88.9","J&F":"87.8","Jaccard (Mean)":"86.7","Speed (FPS)":"29.6"},"uses_additional_data":false},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-1","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (test-dev)","model":"XMem (BL30K, MS)","rank_in_archive_order":7,"of":59,"metrics":{"F-measure (Mean)":"87.0","J&F":"83.7","Jaccard (Mean)":"80.5"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-1","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (test-dev)","model":"XMem (MS)","rank_in_archive_order":9,"of":59,"metrics":{"F-measure (Mean)":"86.4","J&F":"83.1","Jaccard (Mean)":"79.7"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-1","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (test-dev)","model":"XMem (BL30K, 600p)","rank_in_archive_order":12,"of":59,"metrics":{"F-measure (Mean)":"85.8","J&F":"82.5","Jaccard (Mean)":"79.1"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-1","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (test-dev)","model":"XMem (BL30K)","rank_in_archive_order":14,"of":59,"metrics":{"F-measure (Mean)":"84.7","J&F":"81.2","Jaccard (Mean)":"77.6"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-1","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (test-dev)","model":"XMem","rank_in_archive_order":17,"of":59,"metrics":{"F-measure (Mean)":"84.5","J&F":"81.0","Jaccard (Mean)":"77.4"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-1","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (test-dev)","model":"XMem (DAVIS and YouTubeVOS only)","rank_in_archive_order":21,"of":59,"metrics":{"F-measure (Mean)":"83.4","J&F":"79.8","Jaccard (Mean)":"76.3"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2017","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem (BL30K, MS)","rank_in_archive_order":4,"of":81,"metrics":{"F-measure (Mean)":"92.6","J&F":"89.5","Jaccard (Mean)":"86.3"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2017","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem (MS)","rank_in_archive_order":6,"of":81,"metrics":{"F-measure (Mean)":"91.0","J&F":"88.2","Jaccard (Mean)":"85.4"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2017","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem (BL30K)","rank_in_archive_order":11,"of":81,"metrics":{"F-measure (Mean)":"91.4","J&F":"87.7","Jaccard (Mean)":"84.0","Speed (FPS)":"22.6"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2017","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem","rank_in_archive_order":17,"of":81,"metrics":{"F-measure (Mean)":"89.5","J&F":"86.2","Jaccard (Mean)":"82.9","Speed (FPS)":"22.6"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2017","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem (DAVIS and YouTubeVOS only)","rank_in_archive_order":29,"of":81,"metrics":{"F-measure (Mean)":"87.6","J&F":"84.5","Jaccard (Mean)":"81.4","Speed (FPS)":"22.6"},"uses_additional_data":true},{"leaderboard":"/sota/visual-object-tracking-on-davis-2017","task":"Semi-Supervised Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem (DAVIS only)","rank_in_archive_order":53,"of":81,"metrics":{"F-measure (Mean)":"79.3","J&F":"76.7","Jaccard (Mean)":"74.1","Speed (FPS)":"22.6"},"uses_additional_data":false},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-13","task":"Semi-Supervised Video Object Segmentation","dataset":"Long Video Dataset","model":"XMem","rank_in_archive_order":2,"of":9,"metrics":{"F":"91.6±0.2","J":"88.0±0.2","J&F":"89.8±0.2"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-14","task":"Semi-Supervised Video Object Segmentation","dataset":"Long Video Dataset (3X)","model":"XMem","rank_in_archive_order":1,"of":2,"metrics":{"F":"91.8±0.4","J":"88.2±0.3","J&F":"90.0±0.4"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-21","task":"Semi-Supervised Video Object Segmentation","dataset":"MOSE","model":"XMem","rank_in_archive_order":13,"of":17,"metrics":{"F":"62.0","J":"53.3","J&F":"57.6"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2018","model":"XMem (BL30K, MS)","rank_in_archive_order":2,"of":53,"metrics":{"F-Measure (Seen)":"90.3","F-Measure (Unseen)":"90.2","Jaccard (Seen)":"85.6","Jaccard (Unseen)":"81.7","Overall":"86.9"},"uses_additional_data":true},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2018","model":"XMem (MS)","rank_in_archive_order":3,"of":53,"metrics":{"F-Measure (Seen)":"89.9","F-Measure (Unseen)":"89.9","Jaccard (Seen)":"85.3","Jaccard (Unseen)":"81.7","Overall":"86.7"},"uses_additional_data":true},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2018","model":"XMem (BL30K)","rank_in_archive_order":6,"of":53,"metrics":{"F-Measure (Seen)":"89.8","F-Measure (Unseen)":"89.2","Jaccard (Seen)":"85.1","Jaccard (Unseen)":"80.3","Overall":"86.1","Speed  (FPS)":"22.6"},"uses_additional_data":true},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2018","model":"XMem","rank_in_archive_order":9,"of":53,"metrics":{"F-Measure (Seen)":"89.3","F-Measure (Unseen)":"88.7","Jaccard (Seen)":"84.6","Jaccard (Unseen)":"80.2","Overall":"85.7","Speed  (FPS)":"22.6"},"uses_additional_data":true},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2018","model":"XMem (YouTubeVOS only)","rank_in_archive_order":19,"of":53,"metrics":{"F-Measure (Seen)":"88.5","F-Measure (Unseen)":"87.2","Jaccard (Seen)":"83.7","Jaccard (Unseen)":"78.2","Overall":"84.4","Speed  (FPS)":"22.6"},"uses_additional_data":false},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-18","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2019","model":"XMem (BL30K, MS)","rank_in_archive_order":2,"of":22,"metrics":{"F-Measure (Seen)":"89.8","F-Measure (Unseen)":"89.9","Jaccard (Seen)":"85.5","Jaccard (Unseen)":"81.8","Overall":"86.8"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-18","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2019","model":"XMem (MS)","rank_in_archive_order":4,"of":22,"metrics":{"F-Measure (Seen)":"89.2","F-Measure (Unseen)":"89.8","Jaccard (Seen)":"84.9","Jaccard (Unseen)":"81.8","Overall":"86.4"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-18","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2019","model":"XMem (BL30K)","rank_in_archive_order":8,"of":22,"metrics":{"F-Measure (Seen)":"89.2","F-Measure (Unseen)":"88.8","Jaccard (Seen)":"84.8","Jaccard (Unseen)":"80.3","Overall":"85.8"},"uses_additional_data":true},{"leaderboard":"/sota/semi-supervised-video-object-segmentation-on-18","task":"Semi-Supervised Video Object Segmentation","dataset":"YouTube-VOS 2019","model":"XMem","rank_in_archive_order":14,"of":22,"metrics":{"F-Measure (Seen)":"88.0","F-Measure (Unseen)":"87.1","Jaccard (Seen)":"83.6","Jaccard (Unseen)":"78.5","Overall":"84.3"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-davis-2016","task":"Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem (BL30K, MS)","rank_in_archive_order":2,"of":24,"metrics":{"F-Score":"94.4","J&F":"93.3","Jaccard (Mean)":"92.2"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-davis-2016","task":"Video Object Segmentation","dataset":"DAVIS 2016","model":"XMem","rank_in_archive_order":5,"of":24,"metrics":{"F-Score":"92.7","J&F":"91.5","Jaccard (Mean)":"90.4"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-davis-2017-val","task":"Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem (BLK30K, MS)","rank_in_archive_order":1,"of":17,"metrics":{"F-measure":"92.6","Jaccard":"86.3","Mean Jaccard & F-Measure":"89.5"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-davis-2017-val","task":"Video Object Segmentation","dataset":"DAVIS 2017 (val)","model":"XMem","rank_in_archive_order":2,"of":17,"metrics":{"F-measure":"89.5","Jaccard":"82.9","Mean Jaccard & F-Measure":"86.2"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-davis-2017-test","task":"Video Object Segmentation","dataset":"DAVIS-2017 (test-dev)","model":"XMem (BL30K, MS)","rank_in_archive_order":1,"of":2,"metrics":{"F-measure":"87.0","Jaccard":"80.5","Mean Jaccard & F-Measure":"83.7"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-davis-2017-test","task":"Video Object Segmentation","dataset":"DAVIS-2017 (test-dev)","model":"XMem","rank_in_archive_order":2,"of":2,"metrics":{"F-measure":"84.5","Jaccard":"77.4","Mean Jaccard & F-Measure":"81.0"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-m-3-vos","task":"Video Object Segmentation","dataset":"M$^3$-VOS","model":"XMem","rank_in_archive_order":3,"of":4,"metrics":{"Average IOU":"70.4"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos-1","task":"Video Object Segmentation","dataset":"YouTube-VOS 2018","model":"XMem (BL30K, MS)","rank_in_archive_order":1,"of":17,"metrics":{"F-Measure (Seen)":"90.3","F-Measure (Unseen)":"90.2","Jaccard (Seen)":"85.6","Jaccard (Unseen)":"81.7","Mean Jaccard & F-Measure":"86.9"},"uses_additional_data":false},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos-2019-2","task":"Video Object Segmentation","dataset":"YouTube-VOS 2019","model":"XMem (BL30K,MS)","rank_in_archive_order":1,"of":10,"metrics":{"F-Measure (Seen)":"89.8","F-Measure (Unseen)":"89.9","Jaccard (Seen)":"85.5","Jaccard (Unseen)":"81.8","Mean Jaccard & F-Measure":"86.8"},"uses_additional_data":true},{"leaderboard":"/sota/video-object-segmentation-on-youtube-vos-2019-2","task":"Video Object Segmentation","dataset":"YouTube-VOS 2019","model":"XMem","rank_in_archive_order":2,"of":10,"metrics":{"F-Measure (Seen)":"88.6","F-Measure (Unseen)":"88.6","Jaccard (Seen)":"84.3","Jaccard (Unseen)":"80.3","Mean Jaccard & F-Measure":"85.5"},"uses_additional_data":true}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2207.07115","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2207.07115"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/tianyuan168326/videosemanticcompression-pytorch","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/hkchengrex/XMem","reach":null}],"summary":{"ran_draft_wrong":1,"unverified":2},"by_repo_kind":{"official":{"samples":1,"ran":0,"repositories":1},"listed":{"samples":2,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":2,"samples":[{"code_sha256_prefix":"db0ab43be2a00330","entry":"aggregate","repo":"tianyuan168326/videosemanticcompression-pytorch","repo_kind":"listed","path":"XMem/model/network.py","file_url":"https://github.com/tianyuan168326/videosemanticcompression-pytorch/blob/HEAD/XMem/model/network.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"MISDECLARED","metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"db0ab43be2a00330"}},{"code_sha256_prefix":"8f06a84a09a0476d","entry":"XMem","repo":"tianyuan168326/videosemanticcompression-pytorch","repo_kind":"listed","path":"XMem/model/network.py","file_url":"https://github.com/tianyuan168326/videosemanticcompression-pytorch/blob/HEAD/XMem/model/network.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"8f06a84a09a0476d"}},{"code_sha256_prefix":"dc4acccb7d5772df","entry":"XMem","repo":"hkchengrex/XMem","repo_kind":"official","path":"model/network.py","file_url":"https://github.com/hkchengrex/XMem/blob/HEAD/model/network.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"dc4acccb7d5772df"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}