Handwriting Retrieval on Handwriting In-Domain Set
97.51Accuracy@1QWen3-VL-4B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| QWen3-VL-4BParas(M)=4437.82, Latency(ms)=18.212026.01 | 97.51 | 97.91 | 97.98 | 97.74 | 97.78 | |
| Ours w/o FineTuningFine-tuning=false, Paras(M)=1.29, Latency(ms)=2.892026.01 | 97.38 | 100 | 100 | 100 | 98.06 | |
| OursFine-tuning=true, Paras(M)=1.29, Latency(ms)=2.892026.01 | 97.26 | 100 | 100 | 98.63 | 97.95 | |
| MiniCPM-o 2.6Paras(M)=8099.42, Latency(ms)=125.132026.01 | 96.29 | 96.95 | 97.04 | 96.64 | 96.45 | |
| InternVL3.5-8BParas(M)=8528.32, Latency(ms)=188.522026.01 | 95.14 | 97.07 | 97.4 | 96.2 | 68.22 | |
| GME-Qwen2VL-7BParas(M)=7746.38, Latency(ms)=197.232026.01 | 92.44 | 97.76 | 98.14 | 95.19 | 96.22 | |
| RapidOCRParas(M)=15.00, Latency(ms)=96.102026.01 | 90.98 | 92.93 | 93.25 | 92.1 | 90.81 | |
| GME-Qwen2VL-2BParas(M)=2208.99, Latency(ms)=189.432026.01 | 88.74 | 96.11 | 96.89 | 92.57 | 93.37 | |
| EasyOCRParas(M)=30.10, Latency(ms)=20.332026.01 | 85.98 | 88.72 | 89.48 | 87.67 | 83.96 | |
| Chinese CLIP ViT LargeParas(M)=406.23, Latency(ms)=20.602026.01 | 68.77 | 78.37 | 81.14 | 74.81 | 75.24 | |
| Chinese CLIP ViT HugeParas(M)=957.60, Latency(ms)=20.632026.01 | 67.86 | 75.87 | 78.43 | 73.03 | 72.77 | |
| SigLIP 2 GiantParas(M)=1871.39, Latency(ms)=23.792026.01 | 67.34 | 73.86 | 75.74 | 71.56 | 72.17 | |
| Llama-3.2-11B-VisionParas(M)=10670.22, Latency(ms)=717.802026.01 | 66.69 | 69.6 | 70.3 | 68.27 | 67.54 | |
| Phi-3.5-VisionParas(M)=4146.62, Latency(ms)=854.852026.01 | 66.52 | 68.28 | 68.88 | 67.63 | 66.92 | |
| Florence-2-Large (OCR)Paras(M)=776.47, Latency(ms)=159.092026.01 | 65.75 | 67.63 | 68.21 | 67.73 | 66.16 | |
| ABINetParas(M)=36.86, Latency(ms)=14.212026.01 | 64.85 | 66.21 | 66.45 | 65.64 | 63.65 | |
| PARSeqParas(M)=6.02, Latency(ms)=10.752026.01 | 63.88 | 65.85 | 66.29 | 65.03 | 63.22 | |
| Janus-Pro-7BParas(M)=7420.43, Latency(ms)=572.832026.01 | 63.8 | 65.77 | 66.42 | 65 | 64.14 | |
| SuryaParas(M)=38.42, Latency(ms)=96.702026.01 | 63.31 | 65.19 | 65.78 | 64.45 | 62.7 | |
| SigLIP 2 So400mParas(M)=1135.67, Latency(ms)=27.232026.01 | 60.21 | 66.02 | 67.63 | 64.03 | 65.34 | |
| TrOCRParas(M)=333.92, Latency(ms)=29.742026.01 | 59.83 | 63.01 | 63.75 | 61.63 | 56.48 | |
| Chinese CLIP ViT BaseParas(M)=188.26, Latency(ms)=26.352026.01 | 59.73 | 70.34 | 73.59 | 66.51 | 66.92 | |
| Nomic-Embed-VisionParas(M)=229.68, Latency(ms)=4.032026.01 | 59.1 | 64.57 | 65.48 | 62.57 | 63.24 | |
| SigLIP 2 LargeParas(M)=881.53, Latency(ms)=26.572026.01 | 58.49 | 64.22 | 65.96 | 62.32 | 63.41 | |
| AltCLIPParas(M)=864.19, Latency(ms)=26.852026.01 | 57.26 | 62.31 | 63.72 | 60.62 | 61.39 | |
| MM-EMBEDParas(M)=8175.51, Latency(ms)=145.172026.01 | 53.05 | 59.87 | 61.95 | 57.43 | 58.09 | |
| SigLIP 2 ViT-BParas(M)=375.23, Latency(ms)=27.672026.01 | 45.24 | 54.31 | 56.59 | 50.95 | 52.43 | |
| E5-VParas(M)=8355.28, Latency(ms)=232.672026.01 | 43.32 | 50.54 | 53.93 | 48.67 | 48.61 | |
| DeepSeek-OCRParas(M)=3336.11, Latency(ms)=719.442026.01 | 37.6 | 43.5 | 45.63 | 41.83 | 42.67 | |
| BGE-VisualizedParas(M)=872.91, Latency(ms)=34.132026.01 | 35.72 | 44.51 | 48.3 | 42.02 | 42.22 |