Privacy Classification on VISPR (test)
70Balance AccuracyInternVL-4B
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| InternVL-4BFine-tuned for Privacy=true, Framework=Privacy VLM2024.05 | 70 | — | 68 | 85 | 56 | — | — | — | — | 0.39 | 84 | |
| TinyLLavaFine-tuned for Privacy=true, Framework=Privacy VLM2024.05 | 67 | — | 63 | 85 | 50 | — | — | — | — | 0.35 | 85 | |
| CoAgent2024.05 | 60 | — | 56 | 74 | 46 | — | — | — | — | 0.19 | 74 | |
| InternVL-2BFine-tuned for Privacy=true, Framework=Privacy VLM2024.05 | 60 | — | 39 | 89 | 25 | — | — | — | — | 0.25 | 95 | |
| InternVL-4BFine-tuned for Privacy=false2024.05 | 59 | — | 34 | 90 | 21 | — | — | — | — | 0.24 | 96 | |
| ShareGPT2024.05 | 58 | — | 31 | 92 | 19 | — | — | — | — | 0.23 | 97 | |
| CogVLM2024.05 | 58 | — | 42 | 79 | 28 | — | — | — | — | 0.18 | 87 | |
| LLaVA2024.05 | 58 | — | 31 | 89 | 19 | — | — | — | — | 0.22 | 96 | |
| TinyLLaVAFine-tuned for Privacy=false2024.05 | 57 | — | 30 | 86 | 18 | — | — | — | — | 0.19 | 95 | |
| MoELLaVA2024.05 | 56 | — | 37 | 78 | 25 | — | — | — | — | 0.16 | 88 | |
| Otter2024.05 | 55 | — | 76 | 65 | 91 | — | — | — | — | -0.14 | 18 | |
| BLIP2024.05 | 55 | — | 27 | 84 | 16 | — | — | — | — | 0.16 | 95 | |
| GPT-42024.05 | 54 | — | 15 | 94 | 8 | — | — | — | — | 0.16 | 99 | |
| InternVL-2BFine-tuned for Privacy=false2024.05 | 52 | — | 7 | 91 | 4 | — | — | — | — | 0.1 | 99 | |
| Fuyu2024.05 | 50 | — | 75 | 64 | 94 | — | — | — | — | 0 | 6 | |
| PrivLEXInterpretable=true, Test set size=7970 images2026.01 | 0.8811 | 0.8884 | 0.881 | 0.9107 | 0.9105 | 0.9106 | 0.8514 | 0.8516 | 0.8515 | — | — | |
| PrivxITMInterpretable=true, Test set size=7970 images2026.01 | 0.8633 | 0.8716 | 0.8632 | 0.8974 | 0.8969 | 0.8971 | 0.8288 | 0.8297 | 0.8293 | — | — | |
| PEAKInterpretable=true, Test set size=7970 images2026.01 | 0.8358 | 0.8547 | 0.842 | 0.863 | 0.912 | 0.8868 | 0.8386 | 0.7596 | 0.7971 | — | — |