Image Captioning on SCOCO (SPICE, METEOR, BERTScore)
27SPICESCOCO SOTA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SCOCO SOTA2026.07 | 27 | 33.9 | — | — | |
| AVA-VLM (7B)Protocol=LoRA-tuning, Parameters=7B2026.07 | 25.6 | 35 | 39.2 | 49.6 | |
| GPT4oProtocol=5-shot2026.07 | 24.9 | 39.4 | 37.7 | 80 | |
| Qwen2.5 (Baseline) (7B)Protocol=LoRA-tuning, Parameters=7B2026.07 | 23.3 | 32.6 | 38.1 | 48 | |
| GPT4VProtocol=5-shot2026.07 | 23.1 | 38.7 | 33.5 | 101 | |
| Qwen2.5⋆ (7B)Protocol=LoRA-tuning, Parameters=7B2026.07 | 20.9 | 30.2 | 37.2 | 39.5 | |
| GPT4VProtocol=Zero-shot2026.07 | 18.2 | 33.7 | 28 | 119 | |
| Qwen2.5 (7B)Protocol=Zero-shot, Parameters=7B2026.07 | 17.4 | 34.5 | 31 | 90 | |
| MiniGPT4-v2Protocol=Zero-shot2026.07 | 15.5 | 28.3 | 31.1 | 60 | |
| LLaVA-v1.5 (13B)Protocol=5-shot, Parameters=13B2026.07 | 15.1 | 30.8 | 27.2 | 101 | |
| LLaVA-v1.5 (13B)Protocol=Zero-shot, Parameters=13B2026.07 | 14.6 | 29.8 | 27.8 | 84 | |
| LLaVA-v1.5 (7B)Protocol=Zero-shot, Parameters=7B2026.07 | 12.8 | 28.5 | 26.4 | 82 | |
| Human annotation2026.07 | — | — | — | 49 |