Model-Human Comparison on model-vs-human (test)
0.43Error ConsistencyHumans (avg)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Humans (avg)Subject=Human2026.02 | 0.43 | — | — | |
| OpenCLIP ViT-H-14 Fourier-filteredBackbone=ViT-H-14, Input Transformation=Fourier-filtered2026.02 | 0.38 | — | — | |
| OpenCLIP ViT-H-14 BlurredBackbone=ViT-H-14, Input Transformation=Blurred (sigma = 2.5)2026.02 | 0.37 | — | — | |
| OpenCLIP ViT-H-14 Resized (64x64)Backbone=ViT-H-14, Input Transformation=Resized (64x64)2026.02 | 0.35 | — | — | |
| Imagenmodel type=zero-shot, prompts=1 prompt2023.09 | 0.31 | 99 | 71 | |
| ImagenBackbone=Imagen2026.02 | 0.31 | — | — | |
| CLIPmodel type=zero-shot, prompts=80 prompts2023.09 | 0.28 | 57 | 71 | |
| OpenCLIP ViT-H-14Backbone=ViT-H-142026.02 | 0.28 | — | — | |
| Stable Diffusionmodel type=zero-shot, prompts=1 prompt2023.09 | 0.26 | 93 | 69 | |
| CLIPmodel type=zero-shot, prompts=1 prompt2023.09 | 0.26 | 80 | 55 | |
| ViT-22B-384model type=discriminative, training_data=4B images2023.09 | 0.26 | 87 | 80 | |
| ViT-22B-384Backbone=ViT-22B, Resolution=3842026.02 | 0.26 | — | — | |
| RN-50model type=discriminative, augmentation=diffusion noise2023.09 | 0.24 | 57 | 57 | |
| Partimodel type=zero-shot, prompts=1 prompt2023.09 | 0.23 | 92 | 58 | |
| ViT-Lmodel type=discriminative, training_data=IN-21K2023.09 | 0.21 | 42 | 73 | |
| RN-50model type=discriminative, training_data=IN-1K2023.09 | 0.21 | 21 | 56 | |
| RN-50model type=discriminative, mode=train+eval w/ diffusion noise2023.09 | 0.18 | 78 | 43 |