Hateful Meme Detection on Hateful Memes (val)
81.56AUROCEnsemble
Evaluation results
| Method | Links | ||
|---|---|---|---|
| EnsembleSource=Vilio2020.12 | 81.56 | — | |
| ERNIE-ViL LargeSource=Vilio2020.12 | 78.76 | — | |
| ERNIE-ViL BaseSource=Vilio2020.12 | 78.18 | — | |
| UNITERSource=Vilio2020.12 | 77.75 | — | |
| OSCARSource=Vilio2020.12 | 77.16 | — | |
| VisualBERTSource=Vilio2020.12 | 75.49 | — | |
| Best ensemble modelType=Ours2020.12 | 75.21 | 70.93 | |
| VisualBERT COCOSource=Hateful Memes Baseline2020.12 | 73.97 | — | |
| VisualBERT COCOType=Baseline2020.12 | 73.97 | 65.06 | |
| VILBERTSource=Hateful Memes Baseline2020.12 | 71.13 | — | |
| ViLBERTType=Baseline2020.12 | 71.13 | 62.2 | |
| VisualBERTSource=Hateful Memes Baseline2020.12 | 70.6 | — | |
| VisualBERTType=Baseline2020.12 | 70.6 | 62.1 | |
| VILBERT CCSource=Hateful Memes Baseline2020.12 | 70.07 | — | |
| ViLBERT CCType=Baseline2020.12 | 70.07 | 61.4 | |
| GatedCLIPEncoder=frozen CLIP2026.02 | 66 | 59 | |
| DCPMissing rate η=70%, Image presence (%)=100%, Text presence (%)=30%2024.10 | 60.56 | — | |
| DePTMissing rate η=70%, Image presence (%)=100%, Text presence (%)=30%2024.10 | 60.02 | — | |
| MaPLeMissing rate η=70%, Image presence (%)=100%, Text presence (%)=30%2024.10 | 59.36 | — | |
| MMPMissing rate η=70%, Image presence (%)=100%, Text presence (%)=30%2024.10 | 59.02 | — | |
| CoOpMissing rate η=70%, Image presence (%)=100%, Text presence (%)=30%2024.10 | 58.54 | — | |
| CLIP Baseline (avg)Encoder=frozen CLIP, Fusion Mechanism=simple averaging2026.02 | 49 | 50 |