Meme Generation on MuSE (test)
62.5BERTScore F1Gemini-3-Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini-3-Flashzero-shot=true2026.06 | 62.5 | 46.1 | |
| Claude-Sonnet-4.6zero-shot=true2026.06 | 60.7 | 43.9 | |
| Intent ProjectionBackbone=InternVL3-8B, Strategy=full2026.06 | 58 | 37.8 | |
| Intent ProjectionBackbone=Qwen3-VL-8B, Strategy=full2026.06 | 57.2 | 37.1 | |
| InternVL3-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 56 | 35.7 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 55.1 | 34.8 | |
| InternVL3-8BStrategy=+ SFT + Chain2026.06 | 54.3 | 34.1 | |
| GPT-5.4-minizero-shot=true2026.06 | 54.2 | 36.5 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain2026.06 | 53.4 | 33.3 | |
| InternVL3-8BStrategy=+ SFT2026.06 | 53 | 32.9 | |
| Qwen3-VL-8BStrategy=+ SFT2026.06 | 52.1 | 32 | |
| Intent ProjectionBackbone=LLaVA-OneVision-7B, Strategy=full2026.06 | 51.6 | 32 | |
| InternVL3-8BStrategy=+ CoT2026.06 | 49.9 | 29.4 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain + GRPOflat2026.06 | 49.5 | 30 | |
| Qwen3-VL-8BStrategy=+ CoT2026.06 | 48.5 | 28.3 | |
| InternVL3-8BStrategy=Base2026.06 | 48.2 | 27.6 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain2026.06 | 47.9 | 28.6 | |
| Qwen3-VL-8BStrategy=Base2026.06 | 46.8 | 26.2 | |
| LLaVA-OneVision-7BStrategy=+ SFT2026.06 | 46.5 | 27.4 | |
| LLaVA-OneVision-7BStrategy=+ CoT2026.06 | 43.1 | 24.1 | |
| LLaVA-OneVision-7BStrategy=Base2026.06 | 41.3 | 22 |