AI Paper Atlas

Research map / Benchmarks / ImageNet linear probe

ImageNet linear probe leaderboard

16 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on ImageNet linear probe. Best accuracy: 86.4 (DINOv2 with test-time register, NeurIPS 2025).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

ImageNet linear probe · accuracy

16 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
186.4DINOv2 with test-time register
DINOv2 ViT-L/14
Vision Transformers Don't Need Trained Registers
NeurIPS 2025 · Linear probing results table
Attention and transformers
285.2CLIP L/14-336 + LLM2CLIP
CLIP ViT-L/14-336
LLM2CLIP: Powerful Language Model Unlocks Richer Cross-Modality Representation
AAAI 2026 · Table 3
CLIP and zero-shot
384.1RP-KrossFuse
CLIP + DINOv2
When Kernels Multiply, Clusters Unify: Fusing Embeddings with the Kronecker Product
NeurIPS 2025 · Linear probe image benchmarks
Cross-modal learning
482.73DINOv2 + Ours(all, SE(2))
DINOv2 ViT-L/16
Self-Supervised Learning Based on Transformed Image Reconstruction for Equivariance-Coherent Feature Representation
AAAI 2026 · Table 5
Self-supervised and contrastive
580.3DiGIT (732M)
DINOv2
Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective
NeurIPS 2024 · Table: Linear-probe accuracy
Autoregressive and latent generation
680.1Φeat
DINOv3
Φeat: Physically-Grounded Material Feature Representation
ECCV 2026 · Table 2
Recognition and long-tail
778.9MLIP + LS
ViT-B/16
MLIP: Efficient Multi-Perspective Language-Image Pretraining with Exhaustive Data Utilization
ICML 2024 · Table: Influence of label soft
CLIP and zero-shot
876.9MST 300 epochs (DINO linear probing)
DeiT-S
MST: Masked Self-Supervised Transformer for Visual Representation
NeurIPS 2021 · Table 1
CNN features
975.8ESSL(DINO+Barlow Twins)
ResNet-50
Rethinking Generalizability and Discriminability of Self-Supervised Learning from Evolutionary Game Theory Perspective
IJCV 2025 · Table 2
Self-supervised and contrastive
1075l-DAE
ViT-L
Deconstructing Denoising Diffusion Models for Self-Supervised Learning
ICLR 2025 · Comparison with previous basel
Image restoration
6 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.