| Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time | ✓ Link | 84.63 | Model soups (BASIC-L) | 2022-03-10 |
| PaLI: A Jointly-Scaled Multilingual Language-Image Model | ✓ Link | 84.3 | ViT-e | 2022-09-14 |
| Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time | ✓ Link | 84.22 | Model soups (ViT-G/14) | 2022-03-10 |
| Swin Transformer V2: Scaling Up Capacity and Resolution | ✓ Link | 84.00% | SwinV2-G | 2021-11-18 |
| The effectiveness of MAE pre-pretraining for billion-scale pretraining | ✓ Link | 84.0 | MAWS (ViT-6.5B) | 2023-03-23 |
| Scaling Vision Transformers | ✓ Link | 83.33 | ViT-G/14 | 2021-06-08 |
| The effectiveness of MAE pre-pretraining for billion-scale pretraining | ✓ Link | 83.0 | MAWS (ViT-2B) | 2023-03-23 |
| EVA-02: A Visual Representation for Neon Genesis | ✓ Link | 82.4 | EVA-02-L | 2023-03-20 |
| EVA: Exploring the Limits of Masked Visual Representation Learning at Scale | ✓ Link | 81.6 | EVA | 2022-11-14 |
| MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models | ✓ Link | 81.5 | MOAT-4 (IN-22K pretraining) | 2022-10-04 |
| DINOv3 | ✓ Link | 81.4 | DINOv3 (7B) | 2025-08-13 |
| Revisiting Weakly Supervised Pre-Training of Visual Perception Models | ✓ Link | 81.1 | SWAG (ViT H/14) | 2022-01-20 |
| CoCa: Contrastive Captioners are Image-Text Foundation Models | ✓ Link | 80.7 | CoCa | 2022-05-04 |
| MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models | ✓ Link | 80.6 | MOAT-3 (IN-22K pretraining) | 2022-10-04 |
| Perception Encoder: The best visual embeddings are not at the output of the network | ✓ Link | 80.2 | PE-Core G | 2025-04-17 |
| MoE-ViE: Mixture of Experts Vision Encoder for Efficient Image and Video Understanding | ✓ Link | 80.0 | MoE-ViE-H/14 | 2026-08-18 |
| SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features | ✓ Link | 79.8 | SigLIP 2 (g/16) | 2025-02-20 |
| MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models | ✓ Link | 79.3 | MOAT-2 (IN-22K pretraining) | 2022-10-04 |
| CvT: Introducing Convolutions to Vision Transformers | ✓ Link | 78.8 | CvT-W24 384 | 2021-03-29 |
| MOAT: Alternating Mobile Convolution and Attention Brings Strong Vision Models | ✓ Link | 78.4 | MOAT-1 (IN-22K pretraining) | 2022-10-04 |
| DINOv2: Learning Robust Visual Features without Supervision | ✓ Link | 78.4 | DINOv2 (ViT-g/14) | 2023-04-14 |
| Swin Transformer V2: Scaling Up Capacity and Resolution | ✓ Link | 78.08 | SwinV2-B | 2021-11-18 |
| VOLO: Vision Outlooker for Visual Recognition | ✓ Link | 78 | VOLO-D5 | 2021-06-24 |
| VOLO: Vision Outlooker for Visual Recognition | ✓ Link | 77.8 | VOLO-D4 | 2021-06-24 |
| Sigmoid Loss for Language Image Pre-Training | ✓ Link | 77.2 | SigLIP SO | 2023-03-27 |
| Image Captioners Are Scalable Vision Learners Too | ✓ Link | 76.8 | CapPa L/14 + MAP | 2023-06-13 |
| Going deeper with Image Transformers | ✓ Link | 76.7 | CAIT-M36-448 | 2021-03-31 |
| Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervision | ✓ Link | 76.2 | SEER (RegNet10B) | 2022-02-16 |
| Demystifying CLIP Data | ✓ Link | 76.0 | MetaCLIP ViT-bigG/14 | 2023-09-28 |
| Training data-efficient image transformers & distillation through attention | ✓ Link | 75.2 | DeiT-B Distilled 384 (1000 epochs) | 2020-12-23 |
| ResMLP: Feedforward networks for image classification with data-efficient training | ✓ Link | 74.2 | ResMLP-B24/8 22k | 2021-05-07 |
| Three things everyone should know about Vision Transformers | ✓ Link | 73.9 | ViT-B-36x1 | 2022-03-18 |
| Sparse Autoencoders enable Robust and Interpretable Fine-tuning of CLIP models | ✓ Link | 73.9 | SAE-FT (ViT-B/16) | 2026-05-15 |
| ResMLP: Feedforward networks for image classification with data-efficient training | ✓ Link | 73.4 | ResMLP-B24/8 | 2021-05-07 |
| Sequencer: Deep LSTM for Image Classification | ✓ Link | 73.4 | Sequencer2D-L | 2022-05-04 |
| Anchor-based Robust Finetuning of Vision-Language Models | | 72.8 | ARF | 2024-04-09 |
| Distilling Out-of-Distribution Robustness from Vision-Language Foundation Models | ✓ Link | 71.7 | Discrete Adversarial Distillation (ViT-B, 224) | 2023-11-02 |
| LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference | ✓ Link | 71.4 | LeViT-384 | 2021-04-02 |
| Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision | ✓ Link | 70.1 | ALIGN (extra data) | 2021-02-11 |
| LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference | ✓ Link | 69.9 | LeViT-256 | 2021-04-02 |
| ResMLP: Feedforward networks for image classification with data-efficient training | ✓ Link | 69.8 | ResMLP-S24/16 | 2021-05-07 |
| MobileCLIP: Fast Image-Text Models through Multi-Modal Reinforced Training | ✓ Link | 69.8 | MobileCLIP-B | 2023-11-28 |
| When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations | ✓ Link | 69.6 | ResNet-152x2-SAM | 2021-06-03 |
| Multi-label Cluster Discrimination for Visual Representation Learning | ✓ Link | 68.9 | MLCD ViT-L/14 | 2024-07-24 |
| Retaining Knowledge and Enhancing Long-Text Representations in CLIP through Dual-Teacher Distillation | | 68.8 | LongD-CLIP (ViT-L/14) (extra data) | 2025-06-11 |
| Designing Network Design Spaces | ✓ Link | 68.7 | RegNetY-8.0GF | 2020-03-30 |
| LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference | ✓ Link | 68.7 | LeViT-192 | 2021-04-02 |
| ResNet strikes back: An improved training procedure in timm | ✓ Link | 68.7 | ResNet50 (A1) | 2021-10-01 |
| AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities | ✓ Link | 68.2 | AltCLIP | 2022-11-12 |
| LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference | ✓ Link | 67.5 | LeViT-128 | 2021-04-02 |
| When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations | ✓ Link | 67.5 | ViT-B/16-SAM | 2021-06-03 |
| SmartCLIP: Modular Vision-language Alignment with Identification Guarantees | ✓ Link | 66.6 | SmartCLIP (ViT-L/14) | 2025-07-29 |
| ResMLP: Feedforward networks for image classification with data-efficient training | ✓ Link | 66.0 | ResMLP-S12/16 | 2021-05-07 |
| When Vision Transformers Outperform ResNets without Pre-training or Strong Data Augmentations | ✓ Link | 65.5 | Mixer-B/8-SAM | 2021-06-03 |
| LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference | ✓ Link | 63.9 | LeViT-128S | 2021-04-02 |