{"url":"/sota/image-retrieval-on-roxford-hard","task":{"name":"Image Retrieval","url":"/task/image-retrieval","note":null},"dataset":{"name":"ROxford (Hard)","url":"/dataset/roxford"},"category":"Computer Vision","categories":["Computer Vision"],"category_note":null,"description":"**Image Retrieval** is a fundamental and long-standing computer vision task that involves finding images similar to a given query from a large database. It is often considered a form of fine-grained, instance-level classification. The task is integral to image recognition alongside [classification](/task/image-classification) and [cross-modal retrieval](/task/cross-modal-retrieva). By leveraging visual similarity and other criteria, image retrieval enables users to efficiently discover relevant images, making it a crucial tool in applications such as search and recommendation.\r\n\r\n<span class=\"description-source\">[Extending CLIP for Category-to-image Retrieval in E-commerce](https://arxiv.org/abs/2112.11294)</span>\r\n\r\n<span style=\"color:grey; opacity: 0.6\">( Image credit: [DELF](https://github.com/tensorflow/models/tree/master/research/delf) )</span>","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":["mAP"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"mAP":"higher"}},"counts":{"rows":23,"rows_with_code":23,"rows_with_paper_page":23,"rows_dated":23,"rows_using_additional_data":1},"rows":[{"rank_in_archive_order":1,"model":"SuperGlobal","metrics":{"mAP":"80.2"},"uses_additional_data":false,"paper_date":"2023-08-14","paper":"/paper/global-features-are-all-you-need-for-image","paper_url":"https://arxiv.org/abs/2308.06954v2","paper_title":"Global Features are All You Need for Image Retrieval and Reranking","code":"https://github.com/shihaoshao-gh/superglobal","n_code_links":2,"syntology":{"n_ran":13,"n_unverified":5,"n_samples":18,"n_pointer_only_licence":0}},{"rank_in_archive_order":2,"model":"AMES","metrics":{"mAP":"80"},"uses_additional_data":false,"paper_date":"2024-08-06","paper":"/paper/2408-03282","paper_url":"https://arxiv.org/abs/2408.03282v1","paper_title":"AMES: Asymmetric and Memory-Efficient Similarity Estimation for Instance-level Retrieval","code":"https://github.com/pavelsuma/ames","n_code_links":1,"syntology":{"n_ran":13,"n_unverified":0,"n_samples":13,"n_pointer_only_licence":0}},{"rank_in_archive_order":3,"model":"Hypergraph propagation+community selection","metrics":{"mAP":"73"},"uses_additional_data":false,"paper_date":"2021-12-01","paper":"/paper/hypergraph-propagation-and-community","paper_url":"http://proceedings.neurips.cc/paper/2021/hash/1da546f25222c1ee710cf7e2f7a3ff0c-Abstract.html","paper_title":"Hypergraph Propagation and Community Selection for Objects Retrieval","code":"https://github.com/anguoyuan/Hypergraph-Propagation-and-Community-Selection-for-Objects-Retrieval","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"Token","metrics":{"mAP":"66.57"},"uses_additional_data":false,"paper_date":"2021-12-12","paper":"/paper/learning-token-based-representation-for-image","paper_url":"https://arxiv.org/abs/2112.06159v1","paper_title":"Learning Token-based Representation for Image Retrieval","code":"https://github.com/mcc-wh/token","n_code_links":1,"syntology":null},{"rank_in_archive_order":5,"model":"DELG+ α QE reranking+ RRT reranking","metrics":{"mAP":"64"},"uses_additional_data":false,"paper_date":"2021-03-22","paper":"/paper/instance-level-image-retrieval-using","paper_url":"https://arxiv.org/abs/2103.12236v3","paper_title":"Instance-level Image Retrieval using Reranking Transformers","code":"https://github.com/uvavision/rerankingtransformer","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":1,"n_samples":3,"n_pointer_only_licence":3}},{"rank_in_archive_order":6,"model":"FIRe","metrics":{"mAP":"61.2"},"uses_additional_data":false,"paper_date":"2022-01-31","paper":"/paper/learning-super-features-for-image-retrieval-1","paper_url":"https://arxiv.org/abs/2201.13182v1","paper_title":"Learning Super-Features for Image Retrieval","code":"https://github.com/naver/fire","n_code_links":1,"syntology":null},{"rank_in_archive_order":7,"model":"HOW","metrics":{"mAP":"56.9"},"uses_additional_data":false,"paper_date":"2020-07-26","paper":"/paper/learning-and-aggregating-deep-local","paper_url":"https://arxiv.org/abs/2007.13172v1","paper_title":"Learning and aggregating deep local descriptors for instance-level recognition","code":"https://github.com/gtolias/how","n_code_links":2,"syntology":{"n_ran":1,"n_unverified":2,"n_samples":3,"n_pointer_only_licence":0}},{"rank_in_archive_order":8,"model":"ResNet101+ArcFace GLDv2-train-clean","metrics":{"mAP":"51.6"},"uses_additional_data":false,"paper_date":"2020-04-03","paper":"/paper/google-landmarks-dataset-v2-a-large-scale","paper_url":"https://arxiv.org/abs/2004.01804v2","paper_title":"Google Landmarks Dataset v2 -- A Large-Scale Benchmark for Instance-Level Recognition and Retrieval","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":5,"syntology":null},{"rank_in_archive_order":9,"model":"DELF–HQE+SP","metrics":{"mAP":"50.3"},"uses_additional_data":false,"paper_date":"2016-12-19","paper":"/paper/large-scale-image-retrieval-with-attentive","paper_url":"http://arxiv.org/abs/1612.06321v4","paper_title":"Large-Scale Image Retrieval with Attentive Deep Local Features","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":13,"syntology":{"n_ran":3,"n_unverified":9,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":10,"model":"HesAff–rSIFT–HQE+SP","metrics":{"mAP":"49.7"},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/revisiting-oxford-and-paris-large-scale-image","paper_url":"http://arxiv.org/abs/1803.11285v1","paper_title":"Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":2,"syntology":null},{"rank_in_archive_order":11,"model":"DELF–ASMK*+SP","metrics":{"mAP":"43.1 "},"uses_additional_data":false,"paper_date":"2016-12-19","paper":"/paper/large-scale-image-retrieval-with-attentive","paper_url":"http://arxiv.org/abs/1612.06321v4","paper_title":"Large-Scale Image Retrieval with Attentive Deep Local Features","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":13,"syntology":{"n_ran":3,"n_unverified":9,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":12,"model":"HesAff–rSIFT–HQE","metrics":{"mAP":"41.3"},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/revisiting-oxford-and-paris-large-scale-image","paper_url":"http://arxiv.org/abs/1803.11285v1","paper_title":"Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":2,"syntology":null},{"rank_in_archive_order":13,"model":"R–GeM","metrics":{"mAP":"38.5"},"uses_additional_data":false,"paper_date":"2017-11-03","paper":"/paper/fine-tuning-cnn-image-retrieval-with-no-human","paper_url":"http://arxiv.org/abs/1711.02512v2","paper_title":"Fine-tuning CNN Image Retrieval with No Human Annotation","code":"https://github.com/filipradenovic/cnnimageretrieval-pytorch","n_code_links":14,"syntology":{"n_ran":3,"n_unverified":21,"n_samples":24,"n_pointer_only_licence":2}},{"rank_in_archive_order":14,"model":"HesAff–rSIFT–ASMK*+SP","metrics":{"mAP":"36.7"},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/revisiting-oxford-and-paris-large-scale-image","paper_url":"http://arxiv.org/abs/1803.11285v1","paper_title":"Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":2,"syntology":null},{"rank_in_archive_order":15,"model":"HesAff–rSIFT–ASMK*","metrics":{"mAP":"36.4 "},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/revisiting-oxford-and-paris-large-scale-image","paper_url":"http://arxiv.org/abs/1803.11285v1","paper_title":"Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":2,"syntology":null},{"rank_in_archive_order":16,"model":"HesAff–rSIFT–SMK*+SP","metrics":{"mAP":"35.8 "},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/revisiting-oxford-and-paris-large-scale-image","paper_url":"http://arxiv.org/abs/1803.11285v1","paper_title":"Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":2,"syntology":null},{"rank_in_archive_order":17,"model":"HesAff–rSIFT–SMK*","metrics":{"mAP":"35.4 "},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/revisiting-oxford-and-paris-large-scale-image","paper_url":"http://arxiv.org/abs/1803.11285v1","paper_title":"Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":2,"syntology":null},{"rank_in_archive_order":18,"model":"R–R-MAC","metrics":{"mAP":"32.4 "},"uses_additional_data":false,"paper_date":"2015-11-18","paper":"/paper/particular-object-retrieval-with-integral-max","paper_url":"http://arxiv.org/abs/1511.05879v2","paper_title":"Particular object retrieval with integral max-pooling of CNN activations","code":"https://github.com/almazan/deep-image-retrieval","n_code_links":6,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":19,"model":"Dino","metrics":{"mAP":"24.3"},"uses_additional_data":true,"paper_date":"2021-04-29","paper":"/paper/emerging-properties-in-self-supervised-vision","paper_url":"https://arxiv.org/abs/2104.14294v2","paper_title":"Emerging Properties in Self-Supervised Vision Transformers","code":"https://github.com/facebookresearch/dino","n_code_links":32,"syntology":{"n_ran":5,"n_unverified":15,"n_samples":20,"n_pointer_only_licence":2}},{"rank_in_archive_order":20,"model":"R – [O] –MAC","metrics":{"mAP":"18.0 "},"uses_additional_data":false,"paper_date":"2015-11-18","paper":"/paper/particular-object-retrieval-with-integral-max","paper_url":"http://arxiv.org/abs/1511.05879v2","paper_title":"Particular object retrieval with integral max-pooling of CNN activations","code":"https://github.com/almazan/deep-image-retrieval","n_code_links":6,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":21,"model":"R – [O] –CroW","metrics":{"mAP":"13.3 "},"uses_additional_data":false,"paper_date":"2015-12-13","paper":"/paper/cross-dimensional-weighting-for-aggregated","paper_url":"http://arxiv.org/abs/1512.04065v2","paper_title":"Cross-dimensional Weighting for Aggregated Deep Convolutional Features","code":"https://github.com/yahoo/crow","n_code_links":1,"syntology":null},{"rank_in_archive_order":22,"model":"HesAff–rSIFT–VLAD","metrics":{"mAP":"13.2"},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/revisiting-oxford-and-paris-large-scale-image","paper_url":"http://arxiv.org/abs/1803.11285v1","paper_title":"Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking","code":"https://github.com/tensorflow/models/tree/master/research/delf","n_code_links":2,"syntology":null},{"rank_in_archive_order":23,"model":"R – [O] –SPoC","metrics":{"mAP":"12.4"},"uses_additional_data":false,"paper_date":"2015-10-26","paper":"/paper/aggregating-deep-convolutional-features-for","paper_url":"http://arxiv.org/abs/1510.07493v1","paper_title":"Aggregating Deep Convolutional Features for Image Retrieval","code":"https://github.com/talal579/Deep-image-matching","n_code_links":2,"syntology":null}],"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":10,"rows_with_any_sample_ran":8,"distinct_papers_with_graph_line":8,"distinct_papers_with_any_sample_ran":7,"samples_over_distinct_papers":{"n_ran":40,"n_unverified":57,"n_samples":97,"n_pointer_only_licence":7,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":43,"n_unverified":70,"n_samples":113,"n_pointer_only_licence":7,"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"}}}