Concept Insertion on COCO (test)
67.8Appearance Rate (%)CSAE
Evaluation Results
| Method | Links | |
|---|---|---|
| CSAEModel=Qwen3-VL-4B-Instruct, LLM Judge=Gemini-2.5-Flash, Evaluation Scale=1,000 evaluations, Steering scale=+3us2026.06 | 67.8 | |
| SingleModel=Qwen3-VL-4B-Instruct, LLM Judge=Gemini-2.5-Flash, Evaluation Scale=1,000 evaluations, Steering scale=+3us, Description=Best individual CSAE Level-1 concept2026.06 | 58.4 | |
| MSAEModel=Qwen3-VL-4B-Instruct, LLM Judge=Gemini-2.5-Flash, Evaluation Scale=1,000 evaluations, Steering scale=+3us2026.06 | 40.7 | |
| Rand.Model=Qwen3-VL-4B-Instruct, LLM Judge=Gemini-2.5-Flash, Evaluation Scale=1,000 evaluations, Steering scale=+3us2026.06 | 10.8 | |
| NSModel=Qwen3-VL-4B-Instruct, LLM Judge=Gemini-2.5-Flash, Evaluation Scale=1,000 evaluations, Steering scale=none2026.06 | 10.6 |