AI-generated image detection on Chameleon (ACC, AUC)
93.07AccuracyI2P
Evaluation results
| Method | Links | ||
|---|---|---|---|
| I2PTraining Dataset=ProGAN+SDv1.42026.04 | 93.07 | — | |
| I2PTraining Dataset=SDv1.42026.04 | 92.46 | — | |
| PiND2025.11 | 92.4 | — | |
| I2PTraining Dataset=Average2026.04 | 91.41 | — | |
| MIRRORBackbone=DINOv32026.02 | 90.7 | — | |
| ForgeLensTraining Dataset=ProGAN+SDv1.42026.04 | 90.32 | — | |
| ForgeLensTraining Dataset=SDv1.42026.04 | 90.01 | — | |
| EffortTraining Dataset=ProGAN+SDv1.42026.04 | 89.98 | — | |
| EffortTraining Dataset=SDv1.42026.04 | 89.35 | — | |
| ForgeLensTraining Dataset=Average2026.04 | 89.32 | — | |
| I2PTraining Dataset=ProGAN2026.04 | 88.7 | — | |
| EffortTraining Dataset=Average2026.04 | 87.82 | — | |
| ForgeLensTraining Dataset=ProGAN2026.04 | 87.64 | — | |
| MIRRORBackbone=DINOv22026.02 | 85.4 | — | |
| DDAModel Category=Lightweight Detector2026.03 | 84.82 | — | |
| DDA2025.11 | 84.8 | — | |
| EffortTraining Dataset=ProGAN2026.04 | 84.13 | — | |
| DDA2026.02 | 83.5 | — | |
| Fuzzy Decision TreeModel Category=Fuzzy Decision Tree2026.03 | 82.12 | — | |
| Qwen3-VL-8BModel Category=Multimodal Large Language Model, Protocol=Zero-Shot, Base Component for Ensemble=Yes2026.03 | 81.44 | — | |
| Qwen3-VL-32BModel Category=Multimodal Large Language Model, Protocol=Zero-Shot, Base Component for Ensemble=Yes2026.03 | 81.05 | — | |
| Random ForestModel Category=Ensemble Strategy2026.03 | 80.61 | — | |
| Bagging TreeModel Category=Ensemble Strategy2026.03 | 80.48 | — | |
| VotingModel Category=Ensemble Strategy2026.03 | 79.99 | — | |
| DRCT2026.02 | 79.8 | — | |
| Linear SVMModel Category=Ensemble Strategy2026.03 | 79.8 | — | |
| LightGBMModel Category=Ensemble Strategy2026.03 | 79.58 | — | |
| Logistic RegressionModel Category=Ensemble Strategy2026.03 | 79.27 | — | |
| B-Free2026.02 | 78.3 | — | |
| B-Free2025.11 | 78 | — | |
| Qwen2-VL-7BModel Category=Multimodal Large Language Model, Protocol=Zero-Shot, Base Component for Ensemble=Yes2026.03 | 76.93 | — | |
| Qwen2.5-VL-32BModel Category=Multimodal Large Language Model, Protocol=Zero-Shot2026.03 | 74.48 | — | |
| EfficientFormerTraining Dataset=RealHD (Ours)2026.02 | 71.47 | 0.7811 | |
| DRCT2026.07 | 69.9 | 0.7448 | |
| FerretNetTraining Dataset=ProGAN+SDv1.42026.04 | 69.38 | — | |
| AIGI-Det-CalibBase Detector=Effort, Calibration Protocol=Unsupervised2026.02 | 67.99 | — | |
| FerretNetTraining Dataset=SDv1.42026.04 | 67.92 | — | |
| AIGI-Det-CalibBase Detector=UnivFD, Calibration Protocol=Unsupervised2026.02 | 67.44 | — | |
| AIGI-Det-CalibBase Detector=Effort, Calibration Protocol=Supervised2026.02 | 67.38 | — | |
| AIGI-Det-CalibBase Detector=UnivFD, Calibration Protocol=Supervised2026.02 | 67.28 | — | |
| AIGI-Det-CalibBase Detector=AIDE, Calibration Protocol=Unsupervised2026.02 | 67.27 | — | |
| AIGI-Det-CalibBase Detector=CNNSpot, Calibration Protocol=Unsupervised2026.02 | 67.17 | — | |
| AIGI-Det-CalibBase Detector=CNNSpot, Calibration Protocol=Supervised2026.02 | 67.15 | — | |
| AIGI-Det-CalibBase Detector=AIDE, Calibration Protocol=Supervised2026.02 | 67.08 | — | |
| FerretNetTraining Dataset=Average2026.04 | 66.96 | — | |
| Qwen2.5-VL-7BModel Category=Multimodal Large Language Model, Protocol=Zero-Shot2026.03 | 66.8 | — | |
| RNSIDNet2026.07 | 66.65 | 0.768 | |
| XceptionTraining Dataset=RealHD (Ours)2026.02 | 66.25 | 0.7202 | |
| D3Model Category=Lightweight Detector2026.03 | 65.93 | — | |
| NPRTraining Dataset=ProGAN+SDv1.42026.04 | 65.77 | — | |
| AIDE2026.02 | 65.7 | — | |
| CNN-Spot2025.11 | 65.7 | — | |
| AIDE2026.07 | 64.2 | 0.736 | |
| FerretNetTraining Dataset=ProGAN2026.04 | 63.57 | — | |
| AIGI-Det-CalibBase Detector=FreDect, Calibration Protocol=Supervised2026.02 | 63.52 | — | |
| AIGI-Det-CalibBase Detector=FreDect, Calibration Protocol=Unsupervised2026.02 | 63.48 | — | |
| EffortCalibration Protocol=Original2026.02 | 63.15 | — | |
| AIDECalibration Protocol=Original2026.02 | 63.06 | — | |
| UnivFD2025.11 | 62.8 | — | |
| AIDEModel Category=Lightweight Detector, Base Component for Ensemble=Yes2026.03 | 62.61 | — | |
| AIDE2025.11 | 62.6 | — | |
| AIDETraining dataset=GenImage (SDv1.4 subset)2026.04 | 62.6 | — | |
| Aligned2026.02 | 61.3 | — | |
| DAFTraining dataset=GenImage (SDv1.4 subset)2026.04 | 61.14 | — | |
| GramNetTraining dataset=GenImage (SDv1.4 subset)2026.04 | 60.95 | — | |
| XceptionTraining Dataset=DiffusionForensics2026.02 | 60.87 | 0.6549 | |
| VIB-Net2025.11 | 60.8 | — | |
| ResNet-50Training Dataset=RealHD (Ours)2026.02 | 60.45 | 0.6549 | |
| NPRTraining Dataset=Average2026.04 | 60.4 | — | |
| CNNDetTraining dataset=GenImage (SDv1.4 subset)2026.04 | 60.11 | — | |
| DiffAIGDModel Category=Lightweight Detector2026.03 | 59.83 | — | |
| DIRETraining dataset=GenImage (SDv1.4 subset)2026.04 | 59.71 | — | |
| NPRTraining Dataset=RealHD (Ours)2026.02 | 59.7 | 0.621 | |
| NPRTraining Dataset=DiffusionForensics2026.02 | 59.6 | 0.5637 | |
| FatFormerTraining Dataset=RealHD (Ours)2026.02 | 59.6 | 0.6465 | |
| AIGI-Det-CalibBase Detector=LGrad, Calibration Protocol=Supervised2026.02 | 59.54 | — | |
| EfficientFormerTraining Dataset=GenImage2026.02 | 59.45 | 0.7692 | |
| NPR2025.11 | 59.4 | — | |
| DFFreq2025.11 | 59.4 | — | |
| XceptionTraining Dataset=GenImage2026.02 | 59.36 | 0.4915 | |
| NPRModel Category=Lightweight Detector, Base Component for Ensemble=Yes2026.03 | 59.25 | — | |
| NPRTraining Dataset=GenImage2026.02 | 59.19 | 0.5795 | |
| AIGI-Det-CalibBase Detector=Fusing, Calibration Protocol=Supervised2026.02 | 59.18 | — | |
| AIGI-Det-CalibBase Detector=Fusing, Calibration Protocol=Unsupervised2026.02 | 59.16 | — | |
| UnivFDCalibration Protocol=Original2026.02 | 59.11 | — | |
| AIGI-Det-CalibBase Detector=LNP, Calibration Protocol=Supervised2026.02 | 59.1 | — | |
| SAFE2025.11 | 59.1 | — | |
| AIGI-Det-CalibBase Detector=RINE, Calibration Protocol=Supervised2026.02 | 58.83 | — | |
| FreqNetModel Category=Lightweight Detector2026.03 | 58.78 | — | |
| ResNet-50Training Dataset=GenImage2026.02 | 58.61 | 0.5389 | |
| RINECalibration Protocol=Original2026.02 | 58.54 | — | |
| NPRTraining dataset=GenImage (SDv1.4 subset)2026.04 | 58.13 | — | |
| NPRTraining Dataset=SDv1.42026.04 | 58.13 | — | |
| LGradCalibration Protocol=Original2026.02 | 58.09 | — | |
| PatchCraftModel Category=Lightweight Detector2026.03 | 58.08 | — | |
| FusingCalibration Protocol=Original2026.02 | 58.03 | — | |
| CNNSpotCalibration Protocol=Original2026.02 | 57.91 | — | |
| FreDectCalibration Protocol=Original2026.02 | 57.9 | — | |
| FatFormer2026.02 | 57.8 | — | |
| EfficientFormerTraining Dataset=DiffusionForensics2026.02 | 57.61 | 0.609 |