{"about":{"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.","site":"https://codewithpapers.app","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","syntology":{"site":"https://syntology.ai","developers":"https://syntology.ai/developers","mcp":{"server":"https://syntology.ai/mcp","transport":"streamable-http","server_card":"https://syntology.ai/.well-known/mcp/server-card.json","auth":{"type":"trial token, no account","trial_token":"https://syntology.ai/api/oauth/trial/token","method":"POST","docs":"https://syntology.ai/developers"}},"have":"https://syntology.ai/api/graph/have?x=<method, arXiv id or title> (free, answers coverage only)","paper_base":"https://syntology.ai/paper/","atlas_base":"https://app.syntology.ai/?focus="},"machine_readable":[{"url":"https://codewithpapers.app/llms.txt","what":"the machine catalog: every machine-readable file, counted"},{"url":"https://codewithpapers.app/index/manifest.json","what":"paper-to-code index by arXiv id, with Syntology's counts"},{"url":"https://codewithpapers.app/search/manifest.json","what":"site search index (titles, authors) and its files"},{"url":"https://codewithpapers.app/download","what":"bulk files: Syntology's layer, described there"},{"url":"https://codewithpapers.app/build_manifest.json","what":"the build record: inputs, counts, exclusions, probes"}]},"url":"/paper/convit-improving-vision-transformers-with","title":"ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases","arxiv_id":"2103.10697","date":"2021-03-19","proceeding":null,"authors":["Stéphane d'Ascoli","Hugo Touvron","Matthew Leavitt","Ari Morcos","Giulio Biroli","Levent Sagun"],"abstract":"Convolutional architectures have proven extremely successful for vision tasks. Their hard inductive biases enable sample-efficient learning, but come at the cost of a potentially lower performance ceiling. Vision Transformers (ViTs) rely on more flexible self-attention layers, and have recently outperformed CNNs for image classification. However, they require costly pre-training on large external datasets or distillation from pre-trained convolutional networks. In this paper, we ask the following question: is it possible to combine the strengths of these two architectures while avoiding their respective limitations? To this end, we introduce gated positional self-attention (GPSA), a form of positional self-attention which can be equipped with a ``soft\" convolutional inductive bias. We initialise the GPSA layers to mimic the locality of convolutional layers, then give each attention head the freedom to escape locality by adjusting a gating parameter regulating the attention paid to position versus content information. The resulting convolutional-like ViT architecture, ConViT, outperforms the DeiT on ImageNet, while offering a much improved sample efficiency. We further investigate the role of locality in learning by first quantifying how it is encouraged in vanilla self-attention layers, then analysing how it is escaped in GPSA layers. We conclude by presenting various ablations to better understand the success of the ConViT. Our code and models are released publicly at https://github.com/facebookresearch/convit.","url_abs":"https://arxiv.org/abs/2103.10697v2","url_pdf":"https://arxiv.org/pdf/2103.10697v2.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":"convit-improving-vision-transformers-with","repo_url":"https://github.com/facebookresearch/convit","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/SforAiDl/vformer","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/facebookresearch/vissl","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/lyqcom/convit","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"mindspore","reach":{"status":"unanswered"}},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/rwightman/pytorch-image-models","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/MS-Mind/MS-Code-02/tree/main/configs/convit","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/Mind23-2/MindCode-20","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":{"status":"unanswered"}},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/mindspore-courses/External-Attention-MindSpore/blob/main/model/backbone/ConViT.py","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null},{"paper_slug":"convit-improving-vision-transformers-with","repo_url":"https://github.com/mindspore-ecosystem/mindcv/blob/main/mindcv/models/convit.py","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":{"status":"gone","observed_at":"2026-09-17","how":"tree_404+repo_404"}}],"tasks":[{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"inductive-bias","task_name":"Inductive Bias"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[{"method_slug":"attention","method_name":"Attention"},{"method_slug":"attention-dropout","method_name":"Attention Dropout"},{"method_slug":"convit","method_name":"ConViT"},{"method_slug":"deit","method_name":"DeiT"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"feedforward-network","method_name":"Feedforward Network"},{"method_slug":"gpsa","method_name":"GPSA"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[{"slug":"convit","name":"ConViT","full_name":"ConViT"},{"slug":"gpsa","name":"GPSA","full_name":"Gated Positional Self-Attention"}],"results":[{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"ConViT-B+","rank_in_archive_order":534,"of":1060,"metrics":{"GFLOPs":"30","Number of params":"152M","Top 1 Accuracy":"82.5%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"ConViT-B","rank_in_archive_order":541,"of":1060,"metrics":{"GFLOPs":"17","Number of params":"86M","Top 1 Accuracy":"82.4%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"ConViT-S+","rank_in_archive_order":565,"of":1060,"metrics":{"GFLOPs":"10","Number of params":"48M","Top 1 Accuracy":"82.2%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"ConViT-S","rank_in_archive_order":650,"of":1060,"metrics":{"GFLOPs":"5.4","Number of params":"27M","Top 1 Accuracy":"81.3%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"ConViT-Ti+","rank_in_archive_order":905,"of":1060,"metrics":{"GFLOPs":"2","Number of params":"10M","Top 1 Accuracy":"76.7%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"ConViT-Ti","rank_in_archive_order":990,"of":1060,"metrics":{"GFLOPs":"1","Number of params":"6M","Top 1 Accuracy":"73.1%"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2103.10697","atlas_url":"https://app.syntology.ai/?focus=2103.10697","mcp":null,"developers":"https://syntology.ai/developers","held":true,"reason":"arxiv_id_in_atlas","checked_against":{"atlas_ids":549449,"atlas_ids_sha256":"afd72399dafd9906e9062927f70655952385a3d75488558e1f0484e950ab09b7","read_at":"2026-10-05T01:46:21Z"},"agent_calls":[{"tool":"get_citation_path","arguments":{"paper_1":"2103.10697"},"arguments_in":{"paper_2":"another paper's arXiv id or title"},"call":"get_citation_path(paper_1=\"2103.10697\", paper_2=\"…\")"},{"tool":"get_concepts_for_paper","arguments":{"arxiv_id":"2103.10697"},"call":"get_concepts_for_paper(arxiv_id=\"2103.10697\")"}],"by_repo":[{"repo":"facebookresearch/convit","kind":"official","samples":0,"ran":0,"instrument":0,"unverified":0},{"repo":"SforAiDl/vformer","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0,"harvested_for_other_papers":2,"harvested_for":["2103.13413","2103.15691"],"harvested_for_calls":[{"arxiv_id":"2103.13413","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"2103.13413"},"call":"get_harvested_code_for_paper(arxiv_id=\"2103.13413\")"},{"arxiv_id":"2103.15691","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"2103.15691"},"call":"get_harvested_code_for_paper(arxiv_id=\"2103.15691\")"}]},{"repo":"facebookresearch/vissl","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0,"harvested_for_other_papers":4,"harvested_for":["1603.09246","1803.07728","2106.08254","2202.07603"],"harvested_for_calls":[{"arxiv_id":"1603.09246","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"1603.09246"},"call":"get_harvested_code_for_paper(arxiv_id=\"1603.09246\")"},{"arxiv_id":"1803.07728","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"1803.07728"},"call":"get_harvested_code_for_paper(arxiv_id=\"1803.07728\")"},{"arxiv_id":"2106.08254","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"2106.08254"},"call":"get_harvested_code_for_paper(arxiv_id=\"2106.08254\")"},{"arxiv_id":"2202.07603","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"2202.07603"},"call":"get_harvested_code_for_paper(arxiv_id=\"2202.07603\")"}]},{"repo":"lyqcom/convit","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0},{"repo":"rwightman/pytorch-image-models","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0,"harvested_for_other_papers":1,"harvested_for":["2006.08217"],"harvested_for_calls":[{"arxiv_id":"2006.08217","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"2006.08217"},"call":"get_harvested_code_for_paper(arxiv_id=\"2006.08217\")"}]},{"repo":"MS-Mind/MS-Code-02","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0,"harvested_for_other_papers":1,"harvested_for":["2206.10589"],"harvested_for_calls":[{"arxiv_id":"2206.10589","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"2206.10589"},"call":"get_harvested_code_for_paper(arxiv_id=\"2206.10589\")"}]},{"repo":"Mind23-2/MindCode-20","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0},{"repo":"mindspore-courses/External-Attention-MindSpore","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0},{"repo":"mindspore-ecosystem/mindcv","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0}],"by_repo_harvested_for_is":"harvested_for: the arXiv ids (as Syntology's graph holds them) of every paper whose samples Syntology's graph links to this repository, in arXiv-id order, complete; the page names the first 3 and says how many more. A link only: it does not say the papers are the same work, that one copied the other, or that the code implements this paper's method; harvested_for_calls gives the get_harvested_code_for_paper call for each arXiv-shaped id among them","by_repo_is":"one row per repository in the archive's code links, Syntology's graph links or the samples; samples 0 means no sample is linked to this paper, a statement about Syntology's coverage, not about the repository: with harvested_for_other_papers present, Syntology harvested the repository for that many other papers, and without it, Syntology harvested nothing from it; placed_by_identical_code counts samples whose link records no repository, placed here because identical code was harvested from this repository for another paper; ran counts executions on a synthesized input and instrument counts failures of Syntology's instrument, not of the code"},"code_links_note":null,"arxiv_metadata":null,"syntology_extracted_results":null,"record_sha256":"1428f4a0390a07bee7c2080bdd356e2284b083bb3b13344c7621bd076092f36e","record_changed_at":"2026-09-28","record_changed_at_basis":"first_hashed"}