Meme Classification on HatefulMemes
82.1AccuracyQwen2-VL-7B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen2-VL-7BEvaluation Protocol=RA-HMD2025.02 | 82.1 | 91.1 | — | — | — | — | — | |
| Proposed2026.03 | 81.2 | — | — | — | — | 81 | 79 | |
| LLaVA-1.5-7BEvaluation Protocol=RA-HMD2025.02 | 80.9 | 89.7 | — | — | — | — | — | |
| VPD-55B2025.02 | 80.8 | 89.2 | — | — | — | — | — | |
| [24]2026.03 | 79.9 | — | — | — | — | 80 | 79 | |
| Qwen2-VL-2BEvaluation Protocol=RA-HMD2025.02 | 79.1 | 88.4 | — | — | — | — | — | |
| RGCLTraining=Supervised fine-tuned2025.02 | 78.8 | 87 | — | — | — | — | — | |
| [35]2026.03 | 78.8 | — | — | — | — | — | — | |
| LLaVA-1.5-7BEvaluation Protocol=SFT2025.02 | 78.7 | 85.2 | — | — | — | — | — | |
| Qwen2-VL-7BEvaluation Protocol=SFT2025.02 | 78.6 | 86.3 | — | — | — | — | — | |
| STEMTOXBackbone=PaliGemma2025.08 | 78.4 | — | — | — | — | — | 78.11 | |
| [8]2026.03 | 77.7 | — | — | — | — | — | — | |
| [57]2026.03 | 77.2 | — | — | — | — | — | — | |
| [52]2026.03 | 76.4 | — | — | — | — | — | — | |
| Qwen2-VL-2BEvaluation Protocol=SFT2025.02 | 76.2 | 84 | — | — | — | — | — | |
| HateCLIPperTraining=Supervised fine-tuned2025.02 | 76.1 | 85.5 | — | — | — | — | — | |
| M2KE2025.08 | 75.76 | — | — | — | — | — | 75.52 | |
| ProCap2025.08 | 75.1 | — | — | — | — | — | 74.85 | |
| Fulltuning_mode=full2026.06 | 74.68 | — | — | — | — | — | — | |
| Frozentuning_mode=frozen2026.06 | 73.88 | — | — | — | — | — | — | |
| Adaptertuning_mode=adapter2026.06 | 73.71 | — | — | — | — | — | — | |
| BCAdapter2variant=v22026.06 | 73.46 | — | — | — | — | — | — | |
| [10]2026.03 | 72.98 | — | — | — | — | — | — | |
| PromptHate2025.08 | 72.98 | — | — | — | — | — | 72.24 | |
| BCAdapter1variant=v12026.06 | 72.2 | — | — | — | — | — | — | |
| CLIPTraining=Supervised fine-tuned2025.02 | 72 | 79.8 | — | — | — | — | — | |
| GPT-4o2025.02 | 71.3 | — | — | — | — | — | — | |
| [22]2026.03 | 69.47 | — | — | — | — | — | — | |
| RA-HMDbackbone=Qwen2-VL-7B, fine-tuning set=HarMeme, protocol=RKC2025.02 | 69.3 | 77.1 | — | — | — | — | — | |
| RGCLfine-tuning set=HarMeme2025.02 | 66.9 | 69.9 | — | — | — | — | — | |
| GPT-4omode=low-resourced2025.02 | 66.4 | — | — | — | — | — | — | |
| Few-Shot2025.08 | 66 | — | — | — | — | — | 65.8 | |
| LOREHMmode=low-resourced2025.02 | 65.6 | — | — | — | — | — | — | |
| RA-HMDbackbone=LLaVA-1.5-7B, fine-tuning set=HarMeme, protocol=RKC2025.02 | 65.2 | 74.2 | — | — | — | — | — | |
| Qwen2-VL-7Bfine-tuning set=HarMeme, protocol=SFT + zero-shot2025.02 | 64.1 | 71.1 | — | — | — | — | — | |
| CuPL2024.03 | 64 | — | 51 | 50 | 87 | — | — | |
| Qwen2-VL-7BEvaluation Protocol=4-shot2025.02 | 63.8 | 71.5 | — | — | — | — | — | |
| Qwen2-VL-7BEvaluation Protocol=zero-shot2025.02 | 63.2 | 71.9 | — | — | — | — | — | |
| RA-HMDbackbone=Qwen2-VL-2B, fine-tuning set=HarMeme, protocol=RKC2025.02 | 62.8 | 70.9 | — | — | — | — | — | |
| Ours (Teacher)2024.03 | 61 | — | 66 | 58 | 77 | — | — | |
| PaLI-X2024.03 | 61 | — | 51 | 66 | 42 | — | — | |
| Qwen2-VL-7Bfine-tuning set=HarMeme, protocol=SFT + few-shot2025.02 | 60.6 | 72.3 | — | — | — | — | — | |
| ARMADA - LeuclidModel=BERT-12L, Distillation type=ARMADA - Leuclid2026.03 | 60.4 | — | — | — | — | — | — | |
| ARMADA - LcosineModel=BERT-12L, Distillation type=ARMADA - Lcosine2026.03 | 60.2 | — | — | — | — | — | — | |
| CLIP+MLPLabeler=Human, # Ex.=8.5K2024.03 | 60 | — | 48 | 65 | 38 | — | — | |
| Qwen2-VL-2Bfine-tuning set=HarMeme, protocol=SFT + zero-shot2025.02 | 59.7 | 64.1 | — | — | — | — | — | |
| LLaVA-1.5-7Bfine-tuning set=HarMeme, protocol=SFT + zero-shot2025.02 | 59.4 | 63.8 | — | — | — | — | — | |
| ARMADA - LelementwiseModel=BERT-12L, Distillation type=ARMADA - Lelementwise2026.03 | 59.4 | — | — | — | — | — | — | |
| Qwen2-VL-2BEvaluation Protocol=4-shot2025.02 | 59.1 | 61.7 | — | — | — | — | — | |
| xCLIPLinear probing=true, Backbone=ViT-B/16, Pre-train=IT35M2022.10 | 58.5 | — | — | — | — | — | — | |
| Mod-Hatemode=low-resourced2025.02 | 58 | 64.5 | — | — | — | — | — | |
| LLaVA-1.5-7BEvaluation Protocol=zero-shot2025.02 | 57.6 | 63.7 | — | — | — | — | — | |
| ModHate2025.08 | 57.6 | — | — | — | — | — | 53.88 | |
| LLaVA-1.5-7BEvaluation Protocol=4-shot2025.02 | 57.2 | 63.4 | — | — | — | — | — | |
| LLaVA-1.5-7Bfine-tuning set=HarMeme, protocol=SFT + few-shot2025.02 | 56.4 | 63.1 | — | — | — | — | — | |
| CLIPLinear probing=true, Backbone=ViT-B/16, Pre-train=IT35M2022.10 | 55.9 | — | — | — | — | — | — | |
| nCLIPLinear probing=true, Backbone=ViT-B/16, Pre-train=IT35M2022.10 | 55.6 | — | — | — | — | — | — | |
| undistilledModel=BERT-12L, Distillation type=undistilled2026.03 | 55.4 | — | — | — | — | — | — | |
| Qwen2-VL-2BEvaluation Protocol=zero-shot2025.02 | 54.2 | 64.8 | — | — | — | — | — | |
| Qwen2-VL-2Bfine-tuning set=HarMeme, protocol=SFT + few-shot2025.02 | 53.8 | 61.3 | — | — | — | — | — | |
| CLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 53 | — | — | — | — | — | — | |
| CLIP2024.03 | 53 | — | 57 | 51 | 65 | — | — | |
| nCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 52.7 | — | — | — | — | — | — | |
| xCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 52.5 | — | — | — | — | — | — | |
| Ours (Student)Labeler=MC, # Ex.=7K2024.03 | 52 | — | 56 | 50 | 64 | — | — | |
| BLIP-2zero-shot=true, backbone=FlanT5-XL2023.05 | — | 53 | — | — | — | — | — | |
| BLIP-2zero-shot=true, backbone=FlanT5-XXL2023.05 | — | 52 | — | — | — | — | — | |
| BLIP-2zero-shot=true, backbone=Vicuna-7B2023.05 | — | 50.6 | — | — | — | — | — | |
| BLIP-2zero-shot=true, backbone=Vicuna-13B2023.05 | — | 53.7 | — | — | — | — | — | |
| CLIP-ViT-B/16public data=false2021.12 | — | 63.4 | — | — | — | — | — | |
| CLIP-ViT-B/16 (PMD)public data=true, pretraining dataset=PMD2021.12 | — | 56.6 | — | — | — | — | — | |
| Flamingozero-shot=true, backbone=3B2023.05 | — | 53.7 | — | — | — | — | — | |
| Flamingozero-shot=true, backbone=9B2023.05 | — | 57 | — | — | — | — | — | |
| Flamingozero-shot=true, backbone=80B2023.05 | — | 46.4 | — | — | — | — | — | |
| FLAVApublic data=true2021.12 | — | 76.7 | — | — | — | — | — | |
| InstructBLIPzero-shot=true, backbone=FlanT5-XL2023.05 | — | 56.6 | — | — | — | — | — | |
| InstructBLIPzero-shot=true, backbone=FlanT5-XXL2023.05 | — | 54.1 | — | — | — | — | — | |
| InstructBLIPzero-shot=true, backbone=Vicuna-7B2023.05 | — | 59.6 | — | — | — | — | — | |
| InstructBLIPzero-shot=true, backbone=Vicuna-13B2023.05 | — | 57.5 | — | — | — | — | — | |
| ViLBERTpublic data=true2021.12 | — | 74.1 | — | — | — | — | — | |
| VisualBERTpublic data=true2021.12 | — | 74.1 | — | — | — | — | — |