Image Captioning on MSCOCO 500-sample subset (val)
175.3Caption LengthQwen2.5-VL-7B (Base)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen2.5-VL-7B (Base)Backbone=Qwen2.5-VL-7B, Method=Greedy Decoding2026.01 | 175.3 | 35.6 | 9.9 | 74.4 | 8.52 | 8.7 | |
| Self-Val. (w/ BoN. N=10)Backbone=Qwen2.5-VL-7B, Strategy=Best-of-N Selection, N=102026.01 | 172.8 | 16.6 | 4.1 | 72.7 | 8.94 | 9.13 | |
| Self-Val. (w/ FtA. N=3)Backbone=Qwen2.5-VL-7B, Strategy=Filter-then-Aggregate, N=32026.01 | 156.9 | 8.4 | 3.1 | 70.6 | 8.57 | 8.81 | |
| Self-Val. (w/ FtA. N=3)Backbone=LLaVA-1.5-13B, Strategy=Filter-then-Aggregate, N=32026.01 | 121.7 | 21.2 | 5.1 | 76.7 | 7.96 | 8.32 | |
| Self-Val. (w/ FtA. N=3)Backbone=LLaVA-v1.5-7B, Strategy=Filter-then-Aggregate, N=32026.01 | 115.2 | 22.8 | 5.3 | 76.1 | 7.8 | 8.3 | |
| mPLUG-Owl2-7B (Base)Backbone=mPLUG-Owl2-7B, Method=Greedy Decoding2026.01 | 105.8 | 60.8 | 18.9 | 71.6 | 7.28 | 7.66 | |
| Self-Val. (w/ BoN. N=10)Backbone=mPLUG-Owl2-7B, Strategy=Best-of-N Selection, N=102026.01 | 101.2 | 34.2 | 9.4 | 74.5 | 7.38 | 7.96 | |
| LLaVA-1.5-13B (Base)Backbone=LLaVA-1.5-13B, Method=Greedy Decoding2026.01 | 101 | 45.8 | 13 | 77 | 7.65 | 7.29 | |
| LLaVA-v1.5-7B (Base)Backbone=LLaVA-v1.5-7B, Method=Greedy Decoding2026.01 | 100.6 | 50 | 15.4 | 75.1 | 7.5 | 7.89 | |
| VCDBackbone=LLaVA-v1.5-7B2026.01 | 100.4 | 48.6 | 14.9 | 75.2 | 7.54 | 8.13 | |
| NulluBackbone=LLaVA-v1.5-7B2026.01 | 99.9 | 49.4 | 13.5 | 76.6 | 7.71 | 8.18 | |
| DeCoBackbone=LLaVA-v1.5-7B2026.01 | 99 | 42.4 | 12.9 | 73.3 | 6.96 | 7.66 | |
| Self-Val. (w/ FtA. N=3)Backbone=mPLUG-Owl2-7B, Strategy=Filter-then-Aggregate, N=32026.01 | 98.4 | 13.2 | 4.2 | 70.5 | 7.22 | 7.76 | |
| Self-Val. (w/ BoN. N=10)Backbone=LLaVA-v1.5-7B, Strategy=Best-of-N Selection, N=102026.01 | 98.2 | 28.8 | 7.3 | 76.7 | 7.75 | 8.26 | |
| Self-Val. (w/ BoN. N=10)Backbone=LLaVA-1.5-13B, Strategy=Best-of-N Selection, N=102026.01 | 97.2 | 24 | 6.6 | 77.2 | 7.72 | 8.2 | |
| CGDBackbone=LLaVA-v1.5-7B2026.01 | 89.6 | 37.2 | 11.4 | 78.2 | 7.76 | 8.23 | |
| LessBackbone=LLaVA-v1.5-7B2026.01 | 79 | 38.8 | 12 | 77.6 | 7.64 | 8.07 | |
| HALCBackbone=LLaVA-v1.5-7B2026.01 | 72.4 | 32.2 | 10.9 | 69 | 6.37 | 6.51 |