Meme Generation on MemeReaCon PI-G (test)
52.3BERTScore F1Gemini-3-Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini-3-Flashzero-shot=true2026.06 | 52.3 | 44.9 | |
| Claude-Sonnet-4.6zero-shot=true2026.06 | 47.6 | 40.3 | |
| Intent ProjectionBackbone=InternVL3-8B, Strategy=full2026.06 | 41.3 | 35.6 | |
| InternVL3-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 38.9 | 32.6 | |
| Intent ProjectionBackbone=Qwen3-VL-8B, Strategy=full2026.06 | 37.2 | 32.8 | |
| InternVL3-8BStrategy=+ SFT + Chain2026.06 | 36.4 | 29.9 | |
| GPT-5.4-minizero-shot=true2026.06 | 34.9 | 28.4 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 34.2 | 29.5 | |
| InternVL3-8BStrategy=+ SFT2026.06 | 33.8 | 27.6 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain2026.06 | 31.9 | 26.7 | |
| Qwen3-VL-8BStrategy=+ SFT2026.06 | 29.1 | 24.2 | |
| InternVL3-8BStrategy=+ CoT2026.06 | 28.9 | 23.2 | |
| Intent ProjectionBackbone=LLaVA-OneVision-7B, Strategy=full2026.06 | 25.9 | 22.1 | |
| InternVL3-8BStrategy=Base2026.06 | 25.6 | 20.4 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain + GRPOflat2026.06 | 23.6 | 20.1 | |
| Qwen3-VL-8BStrategy=+ CoT2026.06 | 23.6 | 19.1 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain2026.06 | 21.8 | 18.4 | |
| Qwen3-VL-8BStrategy=Base2026.06 | 19.8 | 16 | |
| LLaVA-OneVision-7BStrategy=+ SFT2026.06 | 19.7 | 16.8 | |
| LLaVA-OneVision-7BStrategy=+ CoT2026.06 | 15.8 | 13.5 | |
| LLaVA-OneVision-7BStrategy=Base2026.06 | 12.7 | 10.9 |