Artifact Detection on ArtiBench 1.0 (test)
63AccuracyInternVL3.5-8B + ArtiAgent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InternVL3.5-8B + ArtiAgentFine-tuning data=ArtiAgent 100K, Task type=Artifact-aware2026.02 | 63 | 62 | |
| Qwen2.5-VL-7B + ArtiAgentFine-tuning data=ArtiAgent 100K, Task type=Artifact-aware2026.02 | 62.7 | 62.7 | |
| GPT-4omode=Zero-shot2026.02 | 61.9 | 60.1 | |
| GPT-5mode=Zero-shot2026.02 | 59.9 | 57.7 | |
| Gemini-2.5-Promode=Zero-shot2026.02 | 58.2 | 57.5 | |
| Qwen2.5-VL-7Bmode=Vanilla / Pre-trained2026.02 | 50.1 | 33.6 | |
| InternVL3.5-8Bmode=Vanilla / Pre-trained2026.02 | 49.8 | 35.7 |