Counter-Perception Discrimination on CP-Bench (dev)
94.3F1 ScoreClaude-3-5-Sonnet
Evaluation Results
| Method | Links | |
|---|---|---|
| Claude-3-5-SonnetSource=Closed-sourced (API)2025.04 | 94.3 | |
| GLM-4vSource=Closed-sourced (API)2025.04 | 88.2 | |
| LLaVA-v1.5-13B + AntidoteStrategy=Antidote (Post-training)2025.04 | 88 | |
| GPT-4v-0409Source=Closed-sourced (API)2025.04 | 86 | |
| LLaVA-Next-Mistral-7B + AntidoteStrategy=Antidote (Post-training)2025.04 | 84.4 | |
| GPT-4o-0806Source=Closed-sourced (API)2025.04 | 84.2 | |
| LLaVA-v1.5-7B + AntidoteStrategy=Antidote (Post-training)2025.04 | 82.9 | |
| GPT-4o-mini-0718Source=Closed-sourced (API)2025.04 | 81.8 | |
| LLaVA-Next-Vicuna-13BSource=Open-sourced2025.04 | 65.1 | |
| InternVL-2-ProSource=Closed-sourced (API)2025.04 | 60.3 | |
| LLaVA-Next-Vicuna-7BSource=Open-sourced2025.04 | 48.7 | |
| InternVL2-8BSource=Open-sourced2025.04 | 47.1 | |
| LLaVA-Next-Mistral-7BCategory=Baseline2025.04 | 43.6 | |
| Cogvlm2-19BSource=Open-sourced2025.04 | 42.8 | |
| Qwen2-VL-72BSource=Open-sourced2025.04 | 42.6 | |
| InternVL2-26BSource=Open-sourced2025.04 | 41.2 | |
| Phi-3-VisionSource=Open-sourced2025.04 | 38.8 | |
| MiniCPM-V2.5-8BSource=Open-sourced2025.04 | 37 | |
| Qwen2-VL-7BSource=Open-sourced2025.04 | 33.7 | |
| LLaVA-v1.5-7B + SeVaStrategy=SeVa2025.04 | 25.4 | |
| InstructBLIP-7BSource=Open-sourced2025.04 | 17.8 | |
| LLaVA-v1.5-7B + HA-DPOStrategy=HA-DPO2025.04 | 13.1 | |
| LLaVA-v1.5-7BCategory=Baseline2025.04 | 12.4 | |
| LLaVA-v1.5-7B + VCDStrategy=VCD2025.04 | 12 | |
| LLaVA-v1.5-13BCategory=Baseline2025.04 | 12 |