{"corpus_id":225039882,"paper_sha":"268d347e8a55b5eb82fb5e7d2f800e33c75ab18a","doi":null,"arxiv_id":"2010.11929","pmid":null,"pmcid":null,"mag_id":3094502228,"dblp_id":"journals/corr/abs-2010-11929","acl_id":null,"title":"An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale","year":2020,"publication_date":"2020-10-22","venue":"International Conference on Learning Representations","journal":{"name":"ArXiv","pages":null,"volume":"abs/2010.11929"},"journal_issn":null,"journal_title":null,"publication_types":["JournalArticle"],"pubmed_pub_types":null,"s2_fields_of_study":["Computer Science"],"reference_count":65,"citation_count":58647,"influential_citation_count":6660,"is_open_access":false,"arxiv_categories":["cs.CV","cs.AI","cs.LG"],"arxiv_license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","arxiv_journal_ref":null,"mesh_headings":null,"chemicals":null,"comments_corrections":null,"source_flags":1,"s2_open_access_pdf_url":null,"s2_open_access_landing_url":null,"s2_open_access_license":null,"s2_open_access_status":null,"pmc_open_access_pdf_url":null,"pmc_open_access_landing_url":null,"pmc_open_access_license":null,"pmc_open_access_status":null,"unpaywall_open_access_pdf_url":null,"unpaywall_open_access_landing_url":null,"unpaywall_open_access_license":null,"unpaywall_open_access_status":null,"abstract":"While the Transformer architecture has become the de-facto standard for natural language processing tasks, its applications to computer vision remain limited. In vision, attention is either applied in conjunction with convolutional networks, or used to replace certain components of convolutional networks while keeping their overall structure in place. We show that this reliance on CNNs is not necessary and a pure transformer applied directly to sequences of image patches can perform very well on image classification tasks. When pre-trained on large amounts of data and transferred to multiple mid-sized or small image recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains excellent results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train.","claims":[{"public_id":"cl_68062a72a1cf5eb41830619aa7f54420","status":"active","text":"Pre-training on large amounts of data and transfer to mid-sized or small image recognition benchmarks yields excellent results for Vision Transformer compared with state-of-the-art convolutional networks.","confidence":0.97,"contributors":[],"url":"https://sah.borca.ai/claims/cl_68062a72a1cf5eb41830619aa7f54420"},{"public_id":"cl_afe344c37358f9bf415efeebbe3c6680","status":"active","text":"Vision Transformer can be applied directly to sequences of image patches and perform very well on image classification tasks.","confidence":0.98,"contributors":[],"url":"https://sah.borca.ai/claims/cl_afe344c37358f9bf415efeebbe3c6680"},{"public_id":"cl_fc49176ca8538df9b6cbb5a48a6ac4fc","status":"active","text":"Vision Transformer requires substantially fewer computational resources to train than state-of-the-art convolutional networks.","confidence":0.96,"contributors":[],"url":"https://sah.borca.ai/claims/cl_fc49176ca8538df9b6cbb5a48a6ac4fc"}],"concepts":[{"public_id":"co_2517092e5a01afdd6b684f9e2031640a","status":"active","name":"pre-training on large amounts of data","description":"An initial training stage on large-scale data before downstream adaptation.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_2517092e5a01afdd6b684f9e2031640a"},{"public_id":"co_39d75a5abb61e61d53340d3ad3cea22a","status":"active","name":"computational resources","description":"The amount of computational effort required to train the model.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_39d75a5abb61e61d53340d3ad3cea22a"},{"public_id":"co_526473fc32474000bb5b1fa0ccb9a6b8","status":"active","name":"image recognition benchmarks","description":"Evaluation datasets used to measure how well the model transfers to smaller or mid-sized image recognition problems.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_526473fc32474000bb5b1fa0ccb9a6b8"},{"public_id":"co_5c0bfddd97023ed8720a76c34a05e13d","status":"active","name":"vision transformer","description":"A transformer-based image recognition model that processes an image as a sequence of patches.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_5c0bfddd97023ed8720a76c34a05e13d"},{"public_id":"co_64794b9e6d82760d20e11809e28830bd","status":"active","name":"image classification tasks","description":"Supervised tasks that assign class labels to images.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_64794b9e6d82760d20e11809e28830bd"},{"public_id":"co_8cacac2fb9879c111352a83c84e2834f","status":"active","name":"image patches","description":"Fixed-size regions of an image used as token-like inputs to the transformer.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_8cacac2fb9879c111352a83c84e2834f"},{"public_id":"co_f29b5a8b8925095e49d9f32f1abf637f","status":"active","name":"convolutional networks","description":"Convolution-based vision models used as baseline comparators in the paper.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_f29b5a8b8925095e49d9f32f1abf637f"}],"external_ids":{"DOI":null,"ArXiv":"2010.11929","PubMed":null,"PubMedCentral":null,"MAG":3094502228,"DBLP":"journals/corr/abs-2010-11929","ACL":null},"open_access":{"is_open_access":true,"pdf_url":"https://arxiv.org/pdf/2010.11929","landing_url":"https://arxiv.org/abs/2010.11929","source":"arxiv","pdf_url_source":"derived_arxiv","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","reason":null},"reference_availability":{"status":"available","references_indexed":true,"full_text_available":true,"full_text_source":"arxiv","count_basis":"semantic_scholar_metadata","extraction_status":"not_applicable","reason":null},"source":{"provider":"episteme2","base_corpus":"semantic_scholar_dump","freshness_mode":"unknown","basis":["semantic_scholar_metadata","postgres_metadata"],"limits":["paper metadata is based on indexed upstream scholarly datasets","claims and concepts are available only for extracted papers","absence of claims or concepts means no extracted graph data is available in this response"],"status":"available","degraded":false,"degraded_reasons":[],"diagnostics":{"status":"available","degraded":false,"degraded_reasons":[],"metadata_status":"available","graph_status":"available","abstract_status":"available"},"source_flags":1},"paper_id":351065,"paper_uid":"5ce15807-7679-4b0b-aa58-91bb28dac60c","canonical_identity":{"paper_id":351065,"paper_uid":"5ce15807-7679-4b0b-aa58-91bb28dac60c","identity_status":"available","lookup_basis":"semantic_scholar_external_id","compatibility_path":"corpus_id"},"url":"https://sah.borca.ai/papers/225039882"}