Meme Generation on MemeReaCon PC-G (test)
66.9BERTScore F1Gemini-3-Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini-3-Flashzero-shot=true2026.06 | 66.9 | 60.4 | |
| Claude-Sonnet-4.6zero-shot=true2026.06 | 64.5 | 57.1 | |
| Intent ProjectionBackbone=InternVL3-8B, Strategy=full2026.06 | 59.8 | 57.2 | |
| GPT-5.4-minizero-shot=true2026.06 | 57.8 | 48.5 | |
| Intent ProjectionBackbone=Qwen3-VL-8B, Strategy=full2026.06 | 57.4 | 54.3 | |
| InternVL3-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 56.2 | 54.1 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 53.5 | 50.9 | |
| InternVL3-8BStrategy=+ SFT + Chain2026.06 | 52.9 | 51 | |
| InternVL3-8BStrategy=+ SFT2026.06 | 50.3 | 48.6 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain2026.06 | 50.1 | 47.6 | |
| Qwen3-VL-8BStrategy=+ SFT2026.06 | 47.3 | 44.9 | |
| Intent ProjectionBackbone=LLaVA-OneVision-7B, Strategy=full2026.06 | 43.6 | 43.1 | |
| InternVL3-8BStrategy=+ CoT2026.06 | 41.8 | 42.5 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain + GRPOflat2026.06 | 40.1 | 39.8 | |
| Qwen3-VL-8BStrategy=+ CoT2026.06 | 38.1 | 35.4 | |
| InternVL3-8BStrategy=Base2026.06 | 37.2 | 38.9 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain2026.06 | 36.2 | 36.7 | |
| LLaVA-OneVision-7BStrategy=+ SFT2026.06 | 33.8 | 34.4 | |
| Qwen3-VL-8BStrategy=Base2026.06 | 33.7 | 31.3 | |
| LLaVA-OneVision-7BStrategy=+ CoT2026.06 | 26.1 | 27.7 | |
| LLaVA-OneVision-7BStrategy=Base2026.06 | 20.2 | 22.5 |