Meme Generation on MET-Meme (test)
58.4BERTScore F1Gemini-3-Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini-3-Flashzero-shot=true2026.06 | 58.4 | 40.7 | |
| Claude-Sonnet-4.6zero-shot=true2026.06 | 55.9 | 38.2 | |
| Intent ProjectionBackbone=InternVL3-8B, Strategy=full2026.06 | 51.7 | 30.8 | |
| Intent ProjectionBackbone=Qwen3-VL-8B, Strategy=full2026.06 | 51 | 30.1 | |
| GPT-5.4-minizero-shot=true2026.06 | 49.8 | 30.6 | |
| InternVL3-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 49.8 | 29.1 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 49.2 | 28.3 | |
| InternVL3-8BStrategy=+ SFT + Chain2026.06 | 48.2 | 27.7 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain2026.06 | 47.6 | 26.9 | |
| InternVL3-8BStrategy=+ SFT2026.06 | 47 | 26.6 | |
| Qwen3-VL-8BStrategy=+ SFT2026.06 | 46.3 | 25.8 | |
| Intent ProjectionBackbone=LLaVA-OneVision-7B, Strategy=full2026.06 | 45.7 | 25.8 | |
| InternVL3-8BStrategy=+ CoT2026.06 | 44.3 | 24.7 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain + GRPOflat2026.06 | 44 | 24.3 | |
| Qwen3-VL-8BStrategy=+ CoT2026.06 | 43 | 23.4 | |
| InternVL3-8BStrategy=Base2026.06 | 42.8 | 23 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain2026.06 | 42.4 | 23 | |
| Qwen3-VL-8BStrategy=Base2026.06 | 41.2 | 21.5 | |
| LLaVA-OneVision-7BStrategy=+ SFT2026.06 | 41 | 22.1 | |
| LLaVA-OneVision-7BStrategy=+ CoT2026.06 | 38.2 | 19.6 | |
| LLaVA-OneVision-7BStrategy=Base2026.06 | 36.7 | 17.9 |