Meme Classification on HarMeme Dataset (test)
85.03AccuracyKID-VLM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| KID-VLMgraph_traversal=Hop 2, scoring_model=MiniLM2024.11 | 85.03 | 84.4 | 83.86 | 86.81 | 92.98 | |
| ExplainHMmode=LLM Debate2024.11 | 84.75 | 77.5 | 80.17 | 75 | 92.05 | |
| RGCL2024.11 | 84.75 | 79.39 | 75.36 | 83.87 | 90.1 | |
| KID-VLMgraph_traversal=Hop 1, scoring_model=MiniLM2024.11 | 84.46 | 83.85 | 83.38 | 86.37 | 92.77 | |
| Pro-Capaugmentation=Target-based Captions2024.11 | 83.9 | 83.18 | 82.62 | 85.38 | 90.97 | |
| HateClipper2024.11 | 83.33 | 79.29 | 70.19 | 91.13 | 91.9 | |
| LLaVazeroshot=true2024.11 | 80.23 | 67.59 | 79.35 | 58.87 | — | |
| PromptHate2024.11 | 78.53 | 78.28 | 80.02 | 82.73 | 90.72 | |
| Mr.Harmaugmentation=Label Rationale2024.11 | 74.29 | 62.76 | 85.83 | 63.31 | — | |
| CLIPFrozen encoders=true2024.11 | 70.62 | 58.01 | 73.79 | 59.55 | 90.8 | |
| CLIPFrozen encoders=false2024.11 | 66.67 | 55.04 | 62.21 | 56.51 | 80.82 |