Papers › Emerging Properties in Self-Supervised Vision Transformers

Emerging Properties in Self-Supervised Vision Transformers

29 Apr 2021ICCV 2021 10arXiv:2104.14294archive 2025-07-28

Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, Armand Joulin

In this paper, we question if self-supervised learning provides new properties to Vision Transformer (ViT) that stand out compared to convolutional networks (convnets). Beyond the fact that adapting self-supervised methods to this architecture works particularly well, we make the following observations: first, self-supervised ViT features contain explicit information about the semantic segmentation of an image, which does not emerge as clearly with supervised ViTs, nor with convnets. Second, these features are also excellent k-NN classifiers, reaching 78.3% top-1 on ImageNet with a small ViT. Our study also underlines the importance of momentum encoder, multi-crop training, and the use of small patches with ViTs. We implement our findings into a simple self-supervised method, called DINO, which we interpret as a form of self-distillation with no labels. We show the synergy between DINO and ViTs by achieving 80.1% top-1 on ImageNet in linear evaluation with ViT-Base.

PaperPDFConference PDFCodeCode Syntology ran

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, Syntology's MCP tool lists every function and class Syntology harvested from this paper and whether it ran (how to connect): get_harvested_code_for_paper(arxiv_id="2104.14294")

Code

Syntology Ran 5 of 20 code samples harvested from 5 repositories linked to this paper; 15 have no recorded run. Of those that ran: 1 ran · violated contract; 4 ran · our draft was wrong.

By repository: community (archive-listed): 19 samples from 5 repositories, 4 ran; 1 identical to code first harvested elsewhere. The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

32 repositories listed; official and paper-mentioned ones first.

facebookresearch/dino officialmentioned in papermentioned on GitHubpytorchApache-2.0 report
Expedit-LargeScale-Vision-Transformer/Expedit-DINO mentioned on GitHubpytorchApache-2.0 report
TRAILab/ST-SLidR mentioned on GitHubpytorch report
adrienangeli/dino mentioned on GitHubpytorchApache-2.0 report
ahmedelmahy/myownvit mentioned on GitHubpytorch report
clemsgrs/hipt mentioned on GitHubpytorch report
computationalpathologygroup/hvit mentioned on GitHubpytorch report
facebookresearch/vissl mentioned on GitHubpytorch report
hasibzunair/peekaboo mentioned on GitHubpytorch report
ipmi-icns-uke/sparsam mentioned on GitHubpytorchMIT report
jmnolte/hccnet mentioned on GitHubpytorch report
kaiko-ai/eva mentioned on GitHubpytorchApache-2.0 report
lightly-ai/lightly mentioned on GitHubpytorch report
manantomar/video-occupancy-models mentioned on GitHubpytorch report
rajatkoner08/oodformer mentioned on GitHubpytorch report
sithu31296/simple-object-tracking mentioned on GitHubpytorchMIT report
ttt496/vit-pytorch mentioned on GitHubpytorch report
valeoai/LOST mentioned on GitHubpytorchNOASSERTION report
valeoai/found mentioned on GitHubpytorchApache-2.0 report
vturrisi/solo-learn mentioned on GitHubpytorch report
waltersimoncini/fungivision mentioned on GitHubpytorchMIT report
woctezuma/steam-DINO mentioned on GitHubMIT report
PaddlePaddle/PASSL paddleApache-2.0 report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

20 samples harvested; 5 ran; 0 honoured the contract we drafted; 15 have no recorded run. Read from Syntology's graph 2026-09-24; that is when this build read the record, not when the samples ran.

1ran · violated contract
4ran · our draft was wrong
15unverified

Licence: 2 of the 20 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from 5 repositories linked to this paper, official or community; each sample names its own and says which. Some samples are identical code Syntology first harvested from another repository; for those, this paper's copy is not located and its licence is not recorded. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself; “our draft was wrong” and “fixture could not drive it” are failures of Syntology's instrument, not of the code.

Each sample ends with its code_sha256, Syntology's identity for that exact code. An agent fetches the stored sample with Syntology's MCP tool get_code(code_sha256="…") (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples from a repository marked official are listed first. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of four reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; or the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar). Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

checkpoint_filter_fn ipmi-icns-uke/sparsam/sparsam/models.py community (archive-listed) ran · our draft was wrong MIT (permissive) · e25a3f7ad14df60b · report
compute_embedding jankrepl/mildlyoverfitted/github_adventures/dino/evaluation.py community (archive-listed) ran · our draft was wrong MIT (permissive) · 469756a4afefa152 · report
get_args_parser clemsgrs/hipt/src/train/classification.py community (archive-listed) ran · our draft was wrong no licence file found · pointer only · 6dfcca5f4b2d0098 · report
get_params_groups ipmi-icns-uke/sparsam/sparsam/helper.py community (archive-listed) ran · our draft was wrong MIT (permissive) · ec22c16f1e0653f7 · report
build_model sithu31296/simple-object-tracking/tracking/clip/model.py community (archive-listed) unverified MIT (permissive) · f3830274a7e18d5e · report
coco_class_index sithu31296/simple-object-tracking/tracking/utils.py community (archive-listed) unverified MIT (permissive) · 1143ac599d89fc39 · report
compute_knn jankrepl/mildlyoverfitted/github_adventures/dino/evaluation.py community (archive-listed) unverified MIT (permissive) · 27951b936651c6e9 · report
filter_classes ipmi-icns-uke/sparsam/sparsam/helper.py community (archive-listed) unverified MIT (permissive) · adcec324a2226ec9 · report
get_large_classes ipmi-icns-uke/sparsam/sparsam/helper.py community (archive-listed) unverified MIT (permissive) · daa26eb67407e2ea · report
iou sithu31296/simple-object-tracking/tracking/sort/matching.py community (archive-listed) unverified MIT (permissive) · 4089a9837aebe640 · report
iou_cost sithu31296/simple-object-tracking/tracking/sort/matching.py community (archive-listed) unverified MIT (permissive) · b71fec04b665a291 · report
knn_classifier alijavidani/local_global_representation_learning/eval_knn.py community (archive-listed) unverified Apache-2.0 (permissive) · 6ecb6791e6a9ebfb · report
letterbox sithu31296/simple-object-tracking/tracking/utils.py community (archive-listed) unverified MIT (permissive) · 128c6d6cc48dcf20 · report
load sithu31296/simple-object-tracking/tracking/clip/clip.py community (archive-listed) unverified MIT (permissive) · acaee8a26cc4ebcc · report
load sithu31296/simple-object-tracking/tracking/dino/dino.py community (archive-listed) unverified MIT (permissive) · f0765c91f6e94bc8 · report
min_cost_matching sithu31296/simple-object-tracking/tracking/sort/matching.py community (archive-listed) unverified MIT (permissive) · 6278dd3b36b3eadc · report
min_max_normalize_tensor ipmi-icns-uke/sparsam/sparsam/utils.py community (archive-listed) unverified MIT (permissive) · 9d228d3aaa15dc45 · report
model_inference ipmi-icns-uke/sparsam/sparsam/utils.py community (archive-listed) unverified MIT (permissive) · 3c5097faa5cd9b8b · report
scale_boxes sithu31296/simple-object-tracking/tracking/utils.py community (archive-listed) unverified MIT (permissive) · 55e839c9dc92868b · report
pair identical code first harvested elsewhere ran · violated contract fingerprinted licence of this copy not recorded · 6ba8cee9f5daea41 · report

Tasks

Copy DetectionImage ClassificationImage RetrievalLinear evaluationSelf-Supervised Image ClassificationSelf-Supervised LearningSemantic SegmentationSingle-object discoveryVideo Object DetectionVideo Object SegmentationVisual Place Recognition

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Image Classification OmniBenchmark DINO Average Top-1 Accuracy 38.9 #8 of 22 Archive leaderboard report
Image Retrieval ROxford (Hard) Dino mAP 24.3 #19 of 23 Archive leaderboard report
Image Retrieval ROxford (Medium) Dino mAP 51.5 #19 of 23 Archive leaderboard report
Image Retrieval RParis (Hard) Dino mAP 51.6 #13 of 23 Archive leaderboard report
Image Retrieval RParis (Medium) Dino mAP 75.3 #13 of 23 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (xcit_medium_24_p8) Number of Params 84M #25 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (xcit_medium_24_p8) Top 1 Accuracy 80.3% #25 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-B/8) Number of Params 80M #27 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-B/8) Top 1 Accuracy 80.1% #27 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-S/8) Number of Params 21M #32 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-S/8) Top 1 Accuracy 79.7% #32 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-B/16) Number of Params 85M #46 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-B/16) Top 1 Accuracy 78.2% #46 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-S/16) Number of Params 21M #56 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ViT-S/16) Top 1 Accuracy 77.0% #56 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ResNet-50) Number of Params 24M #75 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet DINO (ResNet-50) Top 1 Accuracy 75.3% #75 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet (finetuned) DINO (ViT-B/16) Number of Params 85M #48 of 65 Archive leaderboard report
Self-Supervised Image Classification ImageNet (finetuned) DINO (ViT-B/16) Top 1 Accuracy 82.8% #48 of 65 Archive leaderboard report
Video Object Segmentation DAVIS 2017 DINO (ViT-B/8, ImageNet retrain) J&F 71.4 #5 of 5 Archive leaderboard report
Visual Place Recognition 17 Places DINO Recall@1 61.82 #4 of 8 Archive leaderboard report
Visual Place Recognition Baidu Mall DINO Recall@1 48.30 #7 of 8 Archive leaderboard report
Visual Place Recognition Gardens Point DINO Recall@1 78.50 #3 of 7 Archive leaderboard report
Visual Place Recognition Hawkins DINO Recall@1 46.61 #2 of 7 Archive leaderboard report
Visual Place Recognition Laurel Caverns DINO Recall@1 41.07 #2 of 7 Archive leaderboard report
Visual Place Recognition Mid-Atlantic Ridge DINO Recall@1 27.72 #2 of 7 Archive leaderboard report
Visual Place Recognition Nardo-Air DINO Recall@1 57.75 #3 of 7 Archive leaderboard report
Visual Place Recognition Nardo-Air R DINO Recall@1 84.51 #4 of 8 Archive leaderboard report
Visual Place Recognition Oxford RobotCar Dataset DINO Recall@1 15.71 #7 of 7 Archive leaderboard report
Visual Place Recognition Pittsburgh-30k-test DINO Recall@1 70.13 #21 of 22 Archive leaderboard report
Visual Place Recognition St Lucia DINO Recall@1 45.22 #13 of 14 Archive leaderboard report
Visual Place Recognition VP-Air DINO Recall@1 24.02 #4 of 7 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: DINO

Absolute Position EncodingsAdamAttentionBPEDINODense ConnectionsDropoutLabel SmoothingLayer NormalizationLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerResidual ConnectionSoftmaxTransformerVision Transformerk-NN

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections