Parameter value regression on DU-110k 1.0 (test)
0.468Gamma Abs ErrorDU-VLM
Evaluation results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DU-VLM2026.02 | 0.468 | 0.166 | 0.008 | 0.102 | 0.057 | 0.08 | 0.044 | 0.069 | 0.082 | 0.202 | 0.644 | 0.118 | 0 | 0 | 0.186 | 0.105 | |
| Qwen3VL-8B-SFTModel Scale=8B, Supervised Fine-Tuning (SFT)=true2026.02 | 0.486 | 0.172 | 0.009 | 0.107 | 0.054 | 0.078 | 0.057 | 0.093 | 0.082 | 0.21 | 0.839 | 0.174 | 0 | 0 | 0.218 | 0.119 | |
| LLaVA-7B-SFTModel Scale=7B, Supervised Fine-Tuning (SFT)=true2026.02 | 0.621 | 0.218 | 0.012 | 0.142 | 0.067 | 0.104 | 0.081 | 0.132 | 0.166 | 0.408 | 0.668 | 0.127 | 0.047 | 0.023 | 0.237 | 0.165 | |
| GPT-4o2026.02 | 2.316 | 0.832 | 0.314 | 4.355 | 0.401 | 1.061 | 0.375 | 0.706 | 0.45 | 1.711 | 3.578 | 0.61 | 2.97 | 0.995 | 1.486 | 1.467 | |
| Gemini-2.52026.02 | 2.414 | 0.872 | 0.344 | 5.004 | 0.409 | 1.008 | 0.624 | 0.992 | 0.419 | 0.994 | 3.856 | 0.835 | 2.963 | 1 | 1.576 | 1.529 | |
| Qwen3VL-32BModel Scale=32B2026.02 | 2.746 | 0.996 | 0.091 | 1.239 | 0.439 | 1.034 | 0.623 | 0.988 | 0.419 | 1.004 | 3.101 | 0.536 | 2.965 | 1 | 1.483 | 0.971 | |
| Qwen3VL-235BModel Scale=235B2026.02 | 2.746 | 0.996 | 0.114 | 1.613 | 0.435 | 1.005 | 0.534 | 0.875 | 0.434 | 1.232 | 3.452 | 0.656 | 2.963 | 1 | 1.525 | 1.054 | |
| Qwen3VL-8BModel Scale=8B2026.02 | 2.756 | 1 | 0.076 | 1.011 | 0.439 | 1.001 | 0.606 | 0.966 | 0.416 | 0.993 | 2.972 | 0.802 | 2.954 | 0.996 | 1.46 | 0.967 |