3D Object Captioning (Targeted Attack) on Objaverse
27.7AGSCaption Attack
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Caption AttackModel=GPT4Point-2.7B2026.01 | 27.7 | 3.9 | 575 | 2,164 | 6.03 | |
| Vision AttackModel=GPT4Point-2.7B2026.01 | 17.5 | 3.02 | 292 | 1,266 | 4.4 | |
| Caption AttackModel=PointLLM-7B2026.01 | 9.6 | 5.12 | 1,526 | 3,611 | 1.09 | |
| Vision AttackModel=PointLLM-7B2026.01 | 7.1 | 3.51 | 462 | 1,633 | 0.88 | |
| w/o AttackModel=GPT4Point-2.7B2026.01 | 4.2 | 0 | 0 | 0 | — | |
| w/o AttackModel=PointLLM-7B2026.01 | 4 | 0 | 0 | 0 | — |