Meme Abuse Detection on GOAT-Bench 1.0 (test)
75.1Harmfulness AccuracyCROSS-ALIGN+
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CROSS-ALIGN+Backbone=Qwen2.5-VL (7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 75.1 | 73 | 81.2 | 78.6 | 86.2 | 84.8 | 75.3 | 72.6 | 77 | 74.8 | 78.9 | 76.9 | |
| CROSS-ALIGN+Backbone=LLaVA-V1.6 (7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 72.8 | 70.5 | 78.9 | 76.3 | 84 | 82.6 | 72 | 69.1 | 74.9 | 72.5 | 76.5 | 74 | |
| CROSS-ALIGN+Backbone=InternVL2 (7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 71 | 68.6 | 77.5 | 74.7 | 82.3 | 80.9 | 71 | 67.9 | 73.2 | 70.6 | 75 | 72.5 | |
| CROSS-ALIGN+Backbone=DeepSeek-VL-S (7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 69.9 | 67 | 76.5 | 73.7 | 82 | 80.2 | 70.3 | 67 | 72.6 | 69.8 | 74.3 | 71.6 | |
| CROSS-ALIGN+Backbone=Qwen2.5-VL (7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 69.5 | 67.1 | 77 | 74.3 | 83.4 | 81.8 | 71 | 67.8 | 73.1 | 70.5 | 74.8 | 72.3 | |
| CROSS-ALIGN+Backbone=mPLUG-Owl2-Small (7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 69.3 | 66.6 | 76 | 73.3 | 81 | 79.5 | 69 | 65.9 | 71.1 | 68.3 | 73.3 | 70.7 | |
| CROSS-ALIGN+Backbone=MiniGPT-4 (7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 68.6 | 65.9 | 74.6 | 71.9 | 79.5 | 77.8 | 67.2 | 64.1 | 69.5 | 66.6 | 71.9 | 69.3 | |
| CROSS-ALIGN+Backbone=Gemma-VL (7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 68.1 | 65.4 | 74.5 | 71.6 | 80.1 | 78.3 | 68.3 | 65.1 | 70.9 | 67.9 | 72.4 | 69.7 | |
| CROSS-ALIGN+Backbone=LLaVA-V1.6 (7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 67.5 | 65.3 | 74.2 | 71.4 | 80.1 | 78.5 | 69.3 | 66.2 | 71.4 | 69.2 | 72.5 | 70.1 | |
| CROSS-ALIGN+Backbone=InternVL2 (7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 67 | 64.6 | 73.5 | 70.8 | 80.2 | 78.6 | 68.9 | 65.7 | 70 | 67.6 | 71.9 | 69.3 | |
| CROSS-ALIGN+Backbone=mPLUG-Owl2-Small (7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 66.9 | 64 | 73.6 | 70.9 | 79.6 | 77.9 | 68.2 | 65 | 70.1 | 67.4 | 71.7 | 69 | |
| CROSS-ALIGN+Backbone=DeepSeek-VL-S (7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 66.8 | 63.5 | 72.9 | 70.1 | 79 | 77.3 | 67.6 | 64.3 | 69.5 | 66.7 | 70.6 | 67.8 | |
| CROSS-ALIGN+Backbone=BLIP-2 OPT (6.7B), Protocol=Standard In-Context Learning (ICL)2026.02 | 66.7 | 63.9 | 72.8 | 70 | 78.9 | 77.2 | 67.5 | 64.4 | 69 | 66.2 | 70.8 | 68.3 | |
| CROSS-ALIGN+Backbone=MiniGPT-4 (7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 65.2 | 62.6 | 71.8 | 69.1 | 77.5 | 75.9 | 66.5 | 63 | 68.1 | 65.5 | 69.8 | 67 | |
| CROSS-ALIGN+Backbone=Gemma-VL (7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 64.3 | 61.6 | 71.4 | 68.3 | 78.5 | 76.8 | 66 | 62.8 | 68.9 | 65.9 | 69.8 | 67.3 | |
| CROSS-ALIGN+Backbone=BLIP-2 OPT (6.7B), Protocol=Vanilla Zero-Shot Inference2026.02 | 63.9 | 61.2 | 70.5 | 67.4 | 76.1 | 74.3 | 65 | 61.9 | 67.3 | 64.5 | 68.5 | 65.9 | |
| Qwen2.5-VLParameters=7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 63.5 | 61 | 70.2 | 66.7 | 75 | 72.8 | 64 | 61.1 | 66.3 | 63.7 | 67.8 | 65.1 | |
| LLaVA-V1.6Parameters=7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 61.7 | 59.3 | 68.4 | 64.9 | 73.2 | 71.8 | 61.1 | 57.7 | 63.8 | 61.4 | 65.6 | 63 | |
| InternVL2Parameters=7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 59.8 | 57.2 | 66 | 62.4 | 71.1 | 68.5 | 59 | 56 | 61.5 | 58.9 | 63.5 | 60.6 | |
| DeepSeek-VL-SParameters=7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 58.3 | 55.5 | 64 | 60.2 | 69.7 | 66.9 | 57 | 53.7 | 59.9 | 56.9 | 61.8 | 58.6 | |
| mPLUG-Owl2-SmallParameters=7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 58 | 55 | 63.8 | 60.5 | 69.5 | 66.7 | 56.4 | 53.2 | 59.2 | 56.1 | 61.4 | 58.3 | |
| Qwen2.5-VLParameters=7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 57.8 | 55.2 | 64.5 | 61 | 70.2 | 68 | 58.1 | 54.4 | 60 | 56.7 | 62.1 | 59 | |
| MiniGPT-4Parameters=7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 57 | 54.1 | 62.5 | 58.8 | 67.9 | 65.2 | 55 | 51.9 | 57.8 | 54.9 | 60 | 57 | |
| LLaVA-V1.6Parameters=7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 56.9 | 54.7 | 63.1 | 60.5 | 68.8 | 66.2 | 57 | 53.8 | 59.2 | 55.6 | 61 | 58.2 | |
| Gemma-VLParameters=7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 56.5 | 53.6 | 62.7 | 59.1 | 68.5 | 65.4 | 55.8 | 52.7 | 58.6 | 55.4 | 60.4 | 57.2 | |
| InternVL2Parameters=7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 55.9 | 53.1 | 61.2 | 57.9 | 67.1 | 64.4 | 54.7 | 51.2 | 57.5 | 54.8 | 59.3 | 56.3 | |
| mPLUG-Owl2-SmallParameters=7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 55.6 | 52.5 | 61.5 | 58.1 | 67.2 | 64.3 | 54.1 | 51 | 57 | 53.9 | 59.1 | 56 | |
| DeepSeek-VL-SParameters=7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 54.9 | 51.8 | 60.3 | 56.7 | 66.2 | 63.5 | 53.2 | 49.9 | 56.7 | 53.6 | 58.7 | 55.1 | |
| BLIP-2 OPTParameters=6.7B, Protocol=Standard In-Context Learning (ICL)2026.02 | 54.9 | 52.2 | 60.8 | 57 | 66.5 | 63.6 | 53.3 | 50.2 | 56.1 | 53.1 | 58.3 | 55.2 | |
| MiniGPT-4Parameters=7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 53.8 | 51 | 59.6 | 56.3 | 65.4 | 62.9 | 52.4 | 49.3 | 55.7 | 52.4 | 57.4 | 54.4 | |
| Gemma-VLParameters=7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 52.6 | 49.8 | 58.7 | 54.9 | 64.9 | 62.1 | 51.4 | 48.6 | 54.3 | 50.9 | 56.4 | 53.3 | |
| BLIP-2 OPTParameters=6.7B, Protocol=Vanilla Zero-Shot Inference2026.02 | 52.3 | 49.7 | 57.9 | 54.3 | 63.8 | 61.4 | 50.7 | 47.9 | 54 | 50.7 | 55.7 | 52.8 |