{"url":"/sota/image-classification-on-imagenet-v2","task":{"name":"Image Classification","url":"/task/image-classification","note":null},"dataset":{"name":"ImageNet V2","url":"/dataset/imagenet"},"category":"Computer Vision","categories":["Adversarial","Computer Vision"],"category_note":null,"description":"**Image Classification** is a fundamental task in vision recognition that aims to understand and categorize an image as a whole under a specific label. Unlike [object detection](/task/object-detection), which involves classification and location of multiple objects within an image, image classification typically pertains to single-object images. When the classification becomes highly detailed or reaches instance-level, it is often referred to as [image retrieval](/task/image-retrieval), which also involves finding similar images in a large database.\r\n\r\n\r\n<span class=\"description-source\">Source: [Metamorphic Testing for Object Detection Systems ](https://arxiv.org/abs/1912.12162)</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":["Top 1 Accuracy"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"Top 1 Accuracy":"higher"}},"counts":{"rows":33,"rows_with_code":33,"rows_with_paper_page":33,"rows_dated":33,"rows_using_additional_data":12},"rows":[{"rank_in_archive_order":1,"model":"Model soups (BASIC-L)","metrics":{"Top 1 Accuracy":"84.63"},"uses_additional_data":true,"paper_date":"2022-03-10","paper":"/paper/model-soups-averaging-weights-of-multiple","paper_url":"https://arxiv.org/abs/2203.05482v3","paper_title":"Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time","code":"https://github.com/mlfoundations/model-soups","n_code_links":6,"syntology":{"n_ran":5,"n_unverified":12,"n_samples":17,"n_pointer_only_licence":0}},{"rank_in_archive_order":2,"model":"ViT-e","metrics":{"Top 1 Accuracy":"84.3"},"uses_additional_data":false,"paper_date":"2022-09-14","paper":"/paper/pali-a-jointly-scaled-multilingual-language","paper_url":"https://arxiv.org/abs/2209.06794v4","paper_title":"PaLI: A Jointly-Scaled Multilingual Language-Image Model","code":"https://github.com/google-research/big_vision","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":2,"n_samples":4,"n_pointer_only_licence":0}},{"rank_in_archive_order":3,"model":"Model soups (ViT-G/14)","metrics":{"Top 1 Accuracy":"84.22"},"uses_additional_data":true,"paper_date":"2022-03-10","paper":"/paper/model-soups-averaging-weights-of-multiple","paper_url":"https://arxiv.org/abs/2203.05482v3","paper_title":"Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time","code":"https://github.com/mlfoundations/model-soups","n_code_links":6,"syntology":{"n_ran":5,"n_unverified":12,"n_samples":17,"n_pointer_only_licence":0}},{"rank_in_archive_order":4,"model":"SwinV2-G","metrics":{"Top 1 Accuracy":"84.00%"},"uses_additional_data":true,"paper_date":"2021-11-18","paper":"/paper/swin-transformer-v2-scaling-up-capacity-and","paper_url":"https://arxiv.org/abs/2111.09883v2","paper_title":"Swin Transformer V2: Scaling Up Capacity and Resolution","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":23,"syntology":{"n_ran":3,"n_unverified":27,"n_samples":30,"n_pointer_only_licence":0}},{"rank_in_archive_order":5,"model":"MAWS (ViT-6.5B)","metrics":{"Top 1 Accuracy":"84.0"},"uses_additional_data":true,"paper_date":"2023-03-23","paper":"/paper/the-effectiveness-of-mae-pre-pretraining-for","paper_url":"https://arxiv.org/abs/2303.13496v3","paper_title":"The effectiveness of MAE pre-pretraining for billion-scale pretraining","code":"https://github.com/facebookresearch/maws","n_code_links":1,"syntology":null},{"rank_in_archive_order":6,"model":"ViT-G/14","metrics":{"Top 1 Accuracy":"83.33"},"uses_additional_data":true,"paper_date":"2021-06-08","paper":"/paper/scaling-vision-transformers","paper_url":"https://arxiv.org/abs/2106.04560v2","paper_title":"Scaling Vision Transformers","code":"https://github.com/google-research/big_vision","n_code_links":1,"syntology":null},{"rank_in_archive_order":7,"model":"MAWS (ViT-2B)","metrics":{"Top 1 Accuracy":"83.0"},"uses_additional_data":true,"paper_date":"2023-03-23","paper":"/paper/the-effectiveness-of-mae-pre-pretraining-for","paper_url":"https://arxiv.org/abs/2303.13496v3","paper_title":"The effectiveness of MAE pre-pretraining for billion-scale pretraining","code":"https://github.com/facebookresearch/maws","n_code_links":1,"syntology":null},{"rank_in_archive_order":8,"model":"MOAT-4 (IN-22K pretraining)","metrics":{"Top 1 Accuracy":"81.5"},"uses_additional_data":false,"paper_date":"2022-10-04","paper":"/paper/moat-alternating-mobile-convolution-and","paper_url":"https://arxiv.org/abs/2210.01820v2","paper_title":"MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models","code":"https://github.com/google-research/deeplab2","n_code_links":2,"syntology":null},{"rank_in_archive_order":9,"model":"SWAG (ViT H/14)","metrics":{"Top 1 Accuracy":"81.1"},"uses_additional_data":true,"paper_date":"2022-01-20","paper":"/paper/revisiting-weakly-supervised-pre-training-of","paper_url":"https://arxiv.org/abs/2201.08371v2","paper_title":"Revisiting Weakly Supervised Pre-Training of Visual Perception Models","code":"https://github.com/facebookresearch/SWAG","n_code_links":2,"syntology":null},{"rank_in_archive_order":10,"model":"MOAT-3 (IN-22K pretraining)","metrics":{"Top 1 Accuracy":"80.6"},"uses_additional_data":false,"paper_date":"2022-10-04","paper":"/paper/moat-alternating-mobile-convolution-and","paper_url":"https://arxiv.org/abs/2210.01820v2","paper_title":"MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models","code":"https://github.com/google-research/deeplab2","n_code_links":2,"syntology":null},{"rank_in_archive_order":11,"model":"MOAT-2 (IN-22K pretraining)","metrics":{"Top 1 Accuracy":"79.3"},"uses_additional_data":false,"paper_date":"2022-10-04","paper":"/paper/moat-alternating-mobile-convolution-and","paper_url":"https://arxiv.org/abs/2210.01820v2","paper_title":"MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models","code":"https://github.com/google-research/deeplab2","n_code_links":2,"syntology":null},{"rank_in_archive_order":12,"model":"MOAT-1 (IN-22K pretraining)","metrics":{"Top 1 Accuracy":"78.4"},"uses_additional_data":false,"paper_date":"2022-10-04","paper":"/paper/moat-alternating-mobile-convolution-and","paper_url":"https://arxiv.org/abs/2210.01820v2","paper_title":"MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models","code":"https://github.com/google-research/deeplab2","n_code_links":2,"syntology":null},{"rank_in_archive_order":13,"model":"SwinV2-B","metrics":{"Top 1 Accuracy":"78.08"},"uses_additional_data":false,"paper_date":"2021-11-18","paper":"/paper/swin-transformer-v2-scaling-up-capacity-and","paper_url":"https://arxiv.org/abs/2111.09883v2","paper_title":"Swin Transformer V2: Scaling Up Capacity and Resolution","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":23,"syntology":{"n_ran":3,"n_unverified":27,"n_samples":30,"n_pointer_only_licence":0}},{"rank_in_archive_order":14,"model":"VOLO-D5","metrics":{"Top 1 Accuracy":"78"},"uses_additional_data":false,"paper_date":"2021-06-24","paper":"/paper/volo-vision-outlooker-for-visual-recognition","paper_url":"https://arxiv.org/abs/2106.13112v2","paper_title":"VOLO: Vision Outlooker for Visual Recognition","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":7,"syntology":{"n_ran":1,"n_unverified":5,"n_samples":6,"n_pointer_only_licence":0}},{"rank_in_archive_order":15,"model":"VOLO-D4","metrics":{"Top 1 Accuracy":"77.8"},"uses_additional_data":false,"paper_date":"2021-06-24","paper":"/paper/volo-vision-outlooker-for-visual-recognition","paper_url":"https://arxiv.org/abs/2106.13112v2","paper_title":"VOLO: Vision Outlooker for Visual Recognition","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":7,"syntology":{"n_ran":1,"n_unverified":5,"n_samples":6,"n_pointer_only_licence":0}},{"rank_in_archive_order":16,"model":"CAIT-M36-448","metrics":{"Top 1 Accuracy":"76.7"},"uses_additional_data":false,"paper_date":"2021-03-31","paper":"/paper/going-deeper-with-image-transformers","paper_url":"https://arxiv.org/abs/2103.17239v2","paper_title":"Going deeper with Image Transformers","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":21,"syntology":{"n_ran":5,"n_unverified":6,"n_samples":11,"n_pointer_only_licence":2}},{"rank_in_archive_order":17,"model":"SEER (RegNet10B)","metrics":{"Top 1 Accuracy":"76.2"},"uses_additional_data":true,"paper_date":"2022-02-16","paper":"/paper/vision-models-are-more-robust-and-fair-when","paper_url":"https://arxiv.org/abs/2202.08360v2","paper_title":"Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervision","code":"https://github.com/facebookresearch/vissl","n_code_links":1,"syntology":null},{"rank_in_archive_order":18,"model":"ResMLP-B24/8 22k","metrics":{"Top 1 Accuracy":"74.2"},"uses_additional_data":true,"paper_date":"2021-05-07","paper":"/paper/resmlp-feedforward-networks-for-image","paper_url":"https://arxiv.org/abs/2105.03404v2","paper_title":"ResMLP: Feedforward networks for image classification with data-efficient training","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":19,"syntology":{"n_ran":2,"n_unverified":5,"n_samples":7,"n_pointer_only_licence":0}},{"rank_in_archive_order":19,"model":"ViT-B-36x1","metrics":{"Top 1 Accuracy":"73.9"},"uses_additional_data":false,"paper_date":"2022-03-18","paper":"/paper/three-things-everyone-should-know-about","paper_url":"https://arxiv.org/abs/2203.09795v1","paper_title":"Three things everyone should know about Vision Transformers","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":8,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":0}},{"rank_in_archive_order":20,"model":"ResMLP-B24/8","metrics":{"Top 1 Accuracy":"73.4"},"uses_additional_data":false,"paper_date":"2021-05-07","paper":"/paper/resmlp-feedforward-networks-for-image","paper_url":"https://arxiv.org/abs/2105.03404v2","paper_title":"ResMLP: Feedforward networks for image classification with data-efficient training","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":19,"syntology":{"n_ran":2,"n_unverified":5,"n_samples":7,"n_pointer_only_licence":0}},{"rank_in_archive_order":21,"model":"Sequencer2D-L","metrics":{"Top 1 Accuracy":"73.4"},"uses_additional_data":false,"paper_date":"2022-05-04","paper":"/paper/sequencer-deep-lstm-for-image-classification","paper_url":"https://arxiv.org/abs/2205.01972v4","paper_title":"Sequencer: Deep LSTM for Image Classification","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":5,"syntology":{"n_ran":4,"n_unverified":5,"n_samples":9,"n_pointer_only_licence":0}},{"rank_in_archive_order":22,"model":"Discrete Adversarial Distillation (ViT-B, 224)","metrics":{"Top 1 Accuracy":"71.7"},"uses_additional_data":false,"paper_date":"2023-11-02","paper":"/paper/distilling-out-of-distribution-robustness-1","paper_url":"https://arxiv.org/abs/2311.01441v2","paper_title":"Distilling Out-of-Distribution Robustness from Vision-Language Foundation Models","code":"https://github.com/lapisrocks/DiscreteAdversarialDistillation","n_code_links":1,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":0}},{"rank_in_archive_order":23,"model":"LeViT-384","metrics":{"Top 1 Accuracy":"71.4"},"uses_additional_data":false,"paper_date":"2021-04-02","paper":"/paper/levit-a-vision-transformer-in-convnet-s","paper_url":"https://arxiv.org/abs/2104.01136v2","paper_title":"LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference","code":"https://github.com/huggingface/transformers","n_code_links":12,"syntology":{"n_ran":23,"n_unverified":7,"n_samples":30,"n_pointer_only_licence":0}},{"rank_in_archive_order":24,"model":"LeViT-256","metrics":{"Top 1 Accuracy":"69.9"},"uses_additional_data":false,"paper_date":"2021-04-02","paper":"/paper/levit-a-vision-transformer-in-convnet-s","paper_url":"https://arxiv.org/abs/2104.01136v2","paper_title":"LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference","code":"https://github.com/huggingface/transformers","n_code_links":12,"syntology":{"n_ran":23,"n_unverified":7,"n_samples":30,"n_pointer_only_licence":0}},{"rank_in_archive_order":25,"model":"ResMLP-S24/16","metrics":{"Top 1 Accuracy":"69.8"},"uses_additional_data":false,"paper_date":"2021-05-07","paper":"/paper/resmlp-feedforward-networks-for-image","paper_url":"https://arxiv.org/abs/2105.03404v2","paper_title":"ResMLP: Feedforward networks for image classification with data-efficient training","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":19,"syntology":{"n_ran":2,"n_unverified":5,"n_samples":7,"n_pointer_only_licence":0}},{"rank_in_archive_order":26,"model":"ResNet-152x2-SAM","metrics":{"Top 1 Accuracy":"69.6"},"uses_additional_data":true,"paper_date":"2021-06-03","paper":"/paper/when-vision-transformers-outperform-resnets","paper_url":"https://arxiv.org/abs/2106.01548v3","paper_title":"When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations","code":"https://github.com/google-research/vision_transformer","n_code_links":2,"syntology":null},{"rank_in_archive_order":27,"model":"LeViT-192","metrics":{"Top 1 Accuracy":"68.7"},"uses_additional_data":false,"paper_date":"2021-04-02","paper":"/paper/levit-a-vision-transformer-in-convnet-s","paper_url":"https://arxiv.org/abs/2104.01136v2","paper_title":"LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference","code":"https://github.com/huggingface/transformers","n_code_links":12,"syntology":{"n_ran":23,"n_unverified":7,"n_samples":30,"n_pointer_only_licence":0}},{"rank_in_archive_order":28,"model":"ResNet50 (A1)","metrics":{"Top 1 Accuracy":"68.7"},"uses_additional_data":false,"paper_date":"2021-10-01","paper":"/paper/resnet-strikes-back-an-improved-training","paper_url":"https://arxiv.org/abs/2110.00476v1","paper_title":"ResNet strikes back: An improved training procedure in timm","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":14,"syntology":{"n_ran":0,"n_unverified":3,"n_samples":3,"n_pointer_only_licence":0}},{"rank_in_archive_order":29,"model":"LeViT-128","metrics":{"Top 1 Accuracy":"67.5"},"uses_additional_data":false,"paper_date":"2021-04-02","paper":"/paper/levit-a-vision-transformer-in-convnet-s","paper_url":"https://arxiv.org/abs/2104.01136v2","paper_title":"LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference","code":"https://github.com/huggingface/transformers","n_code_links":12,"syntology":{"n_ran":23,"n_unverified":7,"n_samples":30,"n_pointer_only_licence":0}},{"rank_in_archive_order":30,"model":"ViT-B/16-SAM","metrics":{"Top 1 Accuracy":"67.5"},"uses_additional_data":true,"paper_date":"2021-06-03","paper":"/paper/when-vision-transformers-outperform-resnets","paper_url":"https://arxiv.org/abs/2106.01548v3","paper_title":"When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations","code":"https://github.com/google-research/vision_transformer","n_code_links":2,"syntology":null},{"rank_in_archive_order":31,"model":"ResMLP-S12/16","metrics":{"Top 1 Accuracy":"66.0"},"uses_additional_data":false,"paper_date":"2021-05-07","paper":"/paper/resmlp-feedforward-networks-for-image","paper_url":"https://arxiv.org/abs/2105.03404v2","paper_title":"ResMLP: Feedforward networks for image classification with data-efficient training","code":"https://github.com/rwightman/pytorch-image-models","n_code_links":19,"syntology":{"n_ran":2,"n_unverified":5,"n_samples":7,"n_pointer_only_licence":0}},{"rank_in_archive_order":32,"model":"Mixer-B/8-SAM","metrics":{"Top 1 Accuracy":"65.5"},"uses_additional_data":true,"paper_date":"2021-06-03","paper":"/paper/when-vision-transformers-outperform-resnets","paper_url":"https://arxiv.org/abs/2106.01548v3","paper_title":"When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations","code":"https://github.com/google-research/vision_transformer","n_code_links":2,"syntology":null},{"rank_in_archive_order":33,"model":"LeViT-128S","metrics":{"Top 1 Accuracy":"63.9"},"uses_additional_data":false,"paper_date":"2021-04-02","paper":"/paper/levit-a-vision-transformer-in-convnet-s","paper_url":"https://arxiv.org/abs/2104.01136v2","paper_title":"LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference","code":"https://github.com/huggingface/transformers","n_code_links":12,"syntology":{"n_ran":23,"n_unverified":7,"n_samples":30,"n_pointer_only_licence":0}}],"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":21,"rows_with_any_sample_ran":20,"distinct_papers_with_graph_line":11,"distinct_papers_with_any_sample_ran":10,"samples_over_distinct_papers":{"n_ran":47,"n_unverified":72,"n_samples":119,"n_pointer_only_licence":2,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":154,"n_unverified":159,"n_samples":313,"n_pointer_only_licence":2,"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"}}}