| DINOv3 | ✓ Link | 55.9 | DINOv3 + linear probe | 2025-08-13 |
| Vision Pretraining for Dense Spatial Perception | ✓ Link | 53.5 | LingBot-Vision ViT-g + linear probe | 2026-07-06 |
| Vision Pretraining for Dense Spatial Perception | ✓ Link | 52.75 | LingBot-Vision ViT-L + linear probe | 2026-07-06 |
| Vision Pretraining for Dense Spatial Perception | ✓ Link | 51.44 | LingBot-Vision ViT-B + linear probe | 2026-07-06 |
| Aggregating Cross-Domain Knowledge via Learnable Tokens for Multi-Teacher Distillation | ✓ Link | 50.3 | ACTok (ViT-L/16) | 2026-09-08 |
| TIPS: Text-Image Pretraining with Spatial Awareness | ✓ Link | 49.7 | TIPS-SO/14 HR + Linear Probe | 2024-10-21 |
| Perception Encoder: The best visual embeddings are not at the output of the network | ✓ Link | 48.9 | PE-Spatial G | 2025-04-17 |
| Vision Transformers Need Registers | ✓ Link | 47.9 | DINOv2 with Registers (ViT-L/14) + Linear Probe | 2023-09-28 |
| Vision Pretraining for Dense Spatial Perception | ✓ Link | 47.01 | LingBot-Vision ViT-S + linear probe | 2026-07-06 |
| SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features | ✓ Link | 45.4 | SigLIP 2 (So/14, 384px) + linear | 2025-02-20 |
| ViT-Up: Faithful Feature Upsampling for Vision Transformers | ✓ Link | 44.73 | ViT-Up + DINOv3-S+ | 2026-06-12 |
| MuM: Multi-View Masked Image Modeling for 3D Vision | ✓ Link | 30.2 | MuM | 2025-11-21 |
| LeVLJEPA: End-to-End Vision-Language Pretraining Without Negatives | ✓ Link | 23.15 | LeVLJEPA ViT-B/16 (linear segmentation head) | 2026-07-01 |