List your company

Optical Illusion Character Recognition on IlluChar (Noise)

98.9L ScoreHuman
-3.33223.20949.7576.291Mar 24, 2026

Evaluation results

MethodLinks
Human2026.0398.910010099.5—
SMSP (Ours)Model=Qwen3-VL-235B-A22B-Instruct2026.0379.7989488.6—
SMSP (Ours)Model=Qwen3-VL-8B-Instruct2026.0379.195.49387.3—
SMSP (Ours)Model=GLM-4.5V2026.0369.997.492.183.5—
SMSP (Ours)Model=Gemini-2.5-Pro2026.0355.991.685.673.8—
Blur with HistogramModel=GLM-4.5V2026.0345.484.7645.4—
Blur with HistogramModel=Gemini-2.5-Pro2026.0345.218.968.944.5—
SMSP (Ours)Model=GPT-5.22026.0344.885.170.662.9—
Blur with HistogramModel=Qwen3-VL-235B-A22B-Instruct2026.0342.9803.142.2—
Blur with HistogramModel=Qwen3-VL-8B-Instruct2026.0342.6661.437.7—
SMSP (Ours)Model=Claude-Sonnet-4.52026.0341.380.665.458.6—
Blur with HistogramModel=GPT-5.22026.0333.458.72.932—
FilteredModel=GLM-4.5V2026.0327.397.151.753.1—
FilteredModel=Gemini-2.5-Pro2026.0326.585.643.747.3—
Blur with HistogramModel=Claude-Sonnet-4.52026.0325.8603.629.1—
FilteredModel=Qwen3-VL-235B-A22B-Instruct2026.0324.473.128.638.9—
FilteredModel=Qwen3-VL-8B-Instruct2026.0320.959.117.130.3—
FilteredModel=GPT-5.22026.0313.964.633.433.1—
FilteredModel=Claude-Sonnet-4.52026.036.4282.611.2—
CoTModel=GPT-5.22026.033.32.926.79.6—
VanillaModel=GPT-5.22026.033.12.9269.3—
GPT-5.22026.033.12.9269.3-79.8
CoTModel=Qwen3-VL-235B-A22B-Instruct2026.031.715.952.919.6—
VanillaModel=Gemini-2.5-Pro2026.031.717.174.425.8—
Gemini-2.5-Pro2026.031.717.174.425.8-69.1
CoTModel=GLM-4.5V2026.031.524.374.727.7—
CoTModel=Claude-Sonnet-4.52026.031.51.901.2—
VanillaModel=Claude-Sonnet-4.52026.031.41.70.31.2—
Claude-Sonnet-4.52026.031.41.70.31.2-86.3
VanillaModel=GLM-4.5V2026.031.326.375.728.5—
GLM-4.5V2026.031.326.375.728.5-66.8
CoTModel=Gemini-2.5-Pro2026.031.21768.924—
CoTModel=Qwen3-VL-8B-Instruct2026.03112.349.917.5—
VanillaModel=Qwen3-VL-235B-A22B-Instruct2026.030.916.455.420—
Qwen3-VL-235B-A22B2026.030.916.455.420-77.7
VanillaModel=Qwen3-VL-8B-Instruct2026.030.611.448.716.7—
Qwen3-VL-8B2026.030.611.448.716.7-76.5