{"url":"/sota/optical-flow-estimation-on-kitti-2015-train","task":{"name":"Optical Flow Estimation","url":"/task/optical-flow-estimation","note":null},"dataset":{"name":"KITTI 2015 (train)","url":"/dataset/kitti"},"category":"Computer Vision","categories":["Computer Vision"],"category_note":null,"description":"**Optical Flow Estimation** is a computer vision task that involves computing the motion of objects in an image or a video sequence. The goal of optical flow estimation is to determine the movement of pixels or features in the image, which can be used for various applications such as object tracking, motion analysis, and video compression.\r\n\r\nApproaches for optical flow estimation include correlation-based, block-matching, feature tracking, energy-based, and more recently gradient-based.\r\n\r\nFurther readings:\r\n\r\n- [Optical Flow Estimation](https://www.cs.toronto.edu/~fleet/research/Papers/flowChapter05.pdf)\r\n- [Performance of Optical Flow Techniques](https://www.cs.toronto.edu/~fleet/research/Papers/ijcv-94.pdf)\r\n\r\nDefinition source: [Devon: Deformable Volume Network for Learning Optical Flow ](https://arxiv.org/abs/1802.07351)\r\n\r\nImage credit: [Optical Flow Estimation](https://www.cs.toronto.edu/~fleet/research/Papers/flowChapter05.pdf)","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["F1-all","EPE"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"F1-all":"higher","EPE":null}},"counts":{"rows":19,"rows_with_code":19,"rows_with_paper_page":19,"rows_dated":19,"rows_using_additional_data":0},"rows":[{"rank_in_archive_order":1,"model":"MEMFOF","metrics":{"EPE":"2.93","F1-all":"9.93"},"uses_additional_data":false,"paper_date":"2025-06-29","paper":"/paper/memfof-high-resolution-training-for-memory","paper_url":"https://arxiv.org/abs/2506.23151v1","paper_title":"MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation","code":"https://github.com/msu-video-group/memfof","n_code_links":1,"syntology":{"n_ran":9,"n_unverified":4,"n_samples":13,"n_pointer_only_licence":0}},{"rank_in_archive_order":2,"model":"DPFlow","metrics":{"EPE":"3.37","F1-all":"11.1"},"uses_additional_data":false,"paper_date":"2025-03-19","paper":"/paper/dpflow-adaptive-optical-flow-estimation-with-1","paper_url":"https://arxiv.org/abs/2503.14880v1","paper_title":"DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework","code":"https://github.com/hmorimitsu/ptlflow","n_code_links":1,"syntology":null},{"rank_in_archive_order":3,"model":"DEQ-Flow","metrics":{"EPE":"3.76","F1-all":"13.0"},"uses_additional_data":false,"paper_date":"2022-04-18","paper":"/paper/deep-equilibrium-optical-flow-estimation","paper_url":"https://arxiv.org/abs/2204.08442v1","paper_title":"Deep Equilibrium Optical Flow Estimation","code":"https://github.com/locuslab/deq-flow","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"RPKNet","metrics":{"EPE":"3.79","F1-all":"13.0"},"uses_additional_data":false,"paper_date":"2024-02-01","paper":"/paper/recurrent-partial-kernel-network-for","paper_url":"https://hmorimitsu.com/publication/2024-aaai-rpknet/","paper_title":"Recurrent Partial Kernel Network for Efficient Optical Flow Estimation","code":"https://github.com/hmorimitsu/ptlflow","n_code_links":1,"syntology":null},{"rank_in_archive_order":5,"model":"FlowFormer","metrics":{"EPE":"4.09","F1-all":"14.7"},"uses_additional_data":false,"paper_date":"2022-03-30","paper":"/paper/flowformer-a-transformer-architecture-for","paper_url":"https://arxiv.org/abs/2203.16194v4","paper_title":"FlowFormer: A Transformer Architecture for Optical Flow","code":"https://github.com/drinkingcoder/FlowFormer-Official","n_code_links":1,"syntology":{"n_ran":1,"n_unverified":6,"n_samples":7,"n_pointer_only_licence":0}},{"rank_in_archive_order":6,"model":"GMFlowNet","metrics":{"EPE":"4.24","F1-all":"15.4"},"uses_additional_data":false,"paper_date":"2022-03-21","paper":"/paper/global-matching-with-overlapping-attention","paper_url":"https://arxiv.org/abs/2203.11335v1","paper_title":"Global Matching with Overlapping Attention for Optical Flow Estimation","code":"https://github.com/xiaofeng94/gmflownet","n_code_links":1,"syntology":{"n_ran":7,"n_unverified":7,"n_samples":14,"n_pointer_only_licence":0}},{"rank_in_archive_order":7,"model":"SeparableFlow","metrics":{"EPE":"4.60","F1-all":"15.9"},"uses_additional_data":false,"paper_date":"2021-01-01","paper":"/paper/separable-flow-learning-motion-cost-volumes","paper_url":"http://openaccess.thecvf.com//content/ICCV2021/html/Zhang_Separable_Flow_Learning_Motion_Cost_Volumes_for_Optical_Flow_Estimation_ICCV_2021_paper.html","paper_title":"Separable Flow: Learning Motion Cost Volumes for Optical Flow Estimation","code":"https://github.com/feihuzhang/separableflow","n_code_links":1,"syntology":null},{"rank_in_archive_order":8,"model":"Ef-RAFT","metrics":{"EPE":"4.83","F1-all":"16.45"},"uses_additional_data":false,"paper_date":"2024-01-01","paper":"/paper/rethinking-raft-for-efficient-optical-flow","paper_url":"https://arxiv.org/abs/2401.00833v1","paper_title":"Rethinking RAFT for Efficient Optical Flow","code":"https://github.com/n3slami/Ef-RAFT","n_code_links":1,"syntology":null},{"rank_in_archive_order":9,"model":"GMA","metrics":{"EPE":"4.69","F1-all":"17.1"},"uses_additional_data":false,"paper_date":"2021-04-06","paper":"/paper/learning-to-estimate-hidden-motions-with","paper_url":"https://arxiv.org/abs/2104.02409v3","paper_title":"Learning to Estimate Hidden Motions with Global Motion Aggregation","code":"https://github.com/open-mmlab/mmflow","n_code_links":2,"syntology":{"n_ran":1,"n_unverified":3,"n_samples":4,"n_pointer_only_licence":1}},{"rank_in_archive_order":10,"model":"RAFT","metrics":{"EPE":"5.04","F1-all":"17.4"},"uses_additional_data":false,"paper_date":"2020-03-26","paper":"/paper/raft-recurrent-all-pairs-field-transforms-for","paper_url":"https://arxiv.org/abs/2003.12039v3","paper_title":"RAFT: Recurrent All-Pairs Field Transforms for Optical Flow","code":"https://github.com/pytorch/vision","n_code_links":17,"syntology":{"n_ran":40,"n_unverified":26,"n_samples":66,"n_pointer_only_licence":17}},{"rank_in_archive_order":11,"model":"CRAFT","metrics":{"EPE":"4.88","F1-all":"17.5"},"uses_additional_data":false,"paper_date":"2022-03-31","paper":"/paper/craft-cross-attentional-flow-transformer-for","paper_url":"https://arxiv.org/abs/2203.16896v1","paper_title":"CRAFT: Cross-Attentional Flow Transformer for Robust Optical Flow","code":"https://github.com/askerlee/craft","n_code_links":1,"syntology":{"n_ran":7,"n_unverified":5,"n_samples":12,"n_pointer_only_licence":12}},{"rank_in_archive_order":12,"model":"RAPIDFlow","metrics":{"EPE":"5.87","F1-all":"17.7"},"uses_additional_data":false,"paper_date":"2024-05-01","paper":"/paper/rapidflow-recurrent-adaptable-pyramids-with","paper_url":"https://hmorimitsu.com/publication/2024-icra-rapidflow/","paper_title":"RAPIDFlow: Recurrent Adaptable Pyramids with Iterative Decoding for Efficient Optical Flow Estimation","code":"https://github.com/hmorimitsu/ptlflow","n_code_links":1,"syntology":null},{"rank_in_archive_order":13,"model":"SCV","metrics":{"EPE":"6.80","F1-all":"19.3"},"uses_additional_data":false,"paper_date":"2021-04-05","paper":"/paper/learning-optical-flow-from-a-few-matches","paper_url":"https://arxiv.org/abs/2104.02166v1","paper_title":"Learning Optical Flow from a Few Matches","code":"https://github.com/zacjiang/scv","n_code_links":1,"syntology":null},{"rank_in_archive_order":14,"model":"MaskFlowNet","metrics":{"F1-all":"23.1"},"uses_additional_data":false,"paper_date":"2020-03-24","paper":"/paper/maskflownet-asymmetric-feature-matching-with","paper_url":"https://arxiv.org/abs/2003.10955v2","paper_title":"MaskFlownet: Asymmetric Feature Matching with Learnable Occlusion Mask","code":"https://github.com/open-mmlab/mmflow","n_code_links":3,"syntology":{"n_ran":0,"n_unverified":11,"n_samples":11,"n_pointer_only_licence":0}},{"rank_in_archive_order":15,"model":"HD3","metrics":{"EPE":"13.17","F1-all":"24.0"},"uses_additional_data":false,"paper_date":"2018-12-15","paper":"/paper/hierarchical-discrete-distribution","paper_url":"http://arxiv.org/abs/1812.06264v3","paper_title":"Hierarchical Discrete Distribution Decomposition for Match Density Estimation","code":"https://github.com/ucbdrive/hd3","n_code_links":2,"syntology":{"n_ran":2,"n_unverified":12,"n_samples":14,"n_pointer_only_licence":0}},{"rank_in_archive_order":16,"model":"VCN","metrics":{"EPE":"8.36","F1-all":"25.1"},"uses_additional_data":false,"paper_date":"2019-12-01","paper":"/paper/volumetric-correspondence-networks-for","paper_url":"http://papers.nips.cc/paper/8367-volumetric-correspondence-networks-for-optical-flow","paper_title":"Volumetric Correspondence Networks for Optical Flow","code":"https://github.com/gengshan-y/VCN","n_code_links":2,"syntology":null},{"rank_in_archive_order":17,"model":"FlowNet2","metrics":{"EPE":"10.08","F1-all":"30.0"},"uses_additional_data":false,"paper_date":"2016-12-06","paper":"/paper/flownet-20-evolution-of-optical-flow","paper_url":"http://arxiv.org/abs/1612.01925v1","paper_title":"FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks","code":"https://github.com/NVIDIA/flownet2-pytorch","n_code_links":12,"syntology":{"n_ran":2,"n_unverified":19,"n_samples":21,"n_pointer_only_licence":3}},{"rank_in_archive_order":18,"model":"FastFlowNet","metrics":{"EPE":"12.24","F1-all":"33.1"},"uses_additional_data":false,"paper_date":"2021-03-08","paper":"/paper/fastflownet-a-lightweight-network-for-fast","paper_url":"https://arxiv.org/abs/2103.04524v2","paper_title":"FastFlowNet: A Lightweight Network for Fast Optical Flow Estimation","code":"https://github.com/ltkong218/FastFlowNet","n_code_links":2,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":2}},{"rank_in_archive_order":19,"model":"PWC-Net","metrics":{"EPE":"10.35","F1-all":"33.7"},"uses_additional_data":false,"paper_date":"2017-09-07","paper":"/paper/pwc-net-cnns-for-optical-flow-using-pyramid","paper_url":"http://arxiv.org/abs/1709.02371v3","paper_title":"PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume","code":"https://github.com/NVlabs/PWC-Net","n_code_links":21,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":3}}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,264 of the 9,581 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9581,"papers_checked":6264,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":3316},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":11,"rows_with_any_sample_ran":10,"distinct_papers_with_graph_line":11,"distinct_papers_with_any_sample_ran":10,"samples_over_distinct_papers":{"n_ran":74,"n_unverified":93,"n_samples":167,"n_pointer_only_licence":38,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":74,"n_unverified":93,"n_samples":167,"n_pointer_only_licence":38,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}