Scene captioning on WHOOPS! (test)
85.82CIDErQwen3-VL-Instruct RGBP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-VL-Instruct RGBPScale=8B2026.04 | 85.82 | 17.02 | |
| ProCapBackbone=OpenLLaMA-3B2026.04 | 24.65 | 6.8 | |
| ProCapBackbone=Vicuna-1.5-7B2026.04 | 24.44 | 6.6 | |
| ProCapBackbone=OPT-2.7B2026.04 | 16.72 | 4.82 | |
| ProCapBackbone=TinyLlama-1.1B2026.04 | 15.23 | 4.57 | |
| Qwen3-VL-InstructScale=8B2026.04 | 13.9 | 13.1 | |
| Qwen3-VL-InstructScale=2B2026.04 | 13.04 | 13.78 | |
| Qwen3-VL-InstructScale=4B2026.04 | 12.72 | 12.88 | |
| FastVLMScale=0.5B2026.04 | 8.72 | 8.06 | |
| FastVLMScale=7B2026.04 | 7.84 | 9.08 | |
| ProCapBackbone=TinyLlama-1.1B2026.04 | 7.48 | 5.15 | |
| ProCapBackbone=OpenLLaMA-3B2026.04 | 5.73 | 6.6 | |
| FastVLMScale=1.5B2026.04 | 4.88 | 7.7 | |
| ProCapBackbone=OPT-2.7B2026.04 | 4.68 | 8.9 | |
| Qwen3-VL-Instruct RGBPScale=8B2026.04 | 4 | 9.42 | |
| Qwen3-VL-InstructScale=8B2026.04 | 3.08 | 12.46 | |
| Qwen3-VL-InstructScale=4B2026.04 | 2.98 | 9.04 | |
| ProCapBackbone=Vicuna-1.5-7B2026.04 | 2.22 | 7.12 | |
| Qwen3-VL-InstructScale=2B2026.04 | 1.86 | 8.94 | |
| FastVLMScale=7B2026.04 | 1.34 | 8.66 | |
| FastVLMScale=0.5B2026.04 | 1.02 | 6.78 | |
| FastVLMScale=1.5B2026.04 | 0.84 | 8.5 |