Web Image Translation on Multilingual Web Image Translation TH-EN
43.2BLEUGPT4.1†
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT4.1†Tuning Strategy=Zero-Shot2026.05 | 43.2 | 96.2 | |
| Gemini2.5 Pro†Tuning Strategy=Zero-Shot2026.05 | 42.1 | 91.1 | |
| VaaWIT on Qwen3 (8B)Backbone=Qwen3 (8B), Tuning Strategy=DSAM + VAA2026.05 | 39.8 | 87.5 | |
| Full Fine-TuningBackbone=Qwen3-VL (8B), Tuning Strategy=Full Fine-Tuning2026.05 | 36.8 | 84.4 | |
| VaaWIT on LLaMA3.1 (8B)Backbone=LLaMA3.1 (8B), Tuning Strategy=DSAM + VAA2026.05 | 33.5 | 76.4 | |
| PP-OCR_Microsoft APIType=Cascaded Model2026.05 | 23.5 | 76.5 | |
| EasyOCR_Google APIType=Cascaded Model2026.05 | 20.3 | 69.8 | |
| Chain-of-ThoughtBackbone=Qwen3-VL (8B), Tuning Strategy=Prompting2026.05 | 18.6 | 71 | |
| Qwen3-VL (32B)Tuning Strategy=Zero-Shot, Parameters=32B2026.05 | 14.3 | 69.9 | |
| Qwen3-VL (8B)Tuning Strategy=Zero-Shot, Parameters=8B2026.05 | 12.2 | 64.8 | |
| LoRABackbone=Qwen3-VL (8B), Tuning Strategy=LoRA (r=8)2026.05 | 10.5 | 62.4 | |
| LLaMA3.2 (90B)Tuning Strategy=Zero-Shot, Parameters=90B2026.05 | 5.5 | 48 | |
| LLaMA3.2 (11B)Tuning Strategy=Zero-Shot, Parameters=11B2026.05 | 3.1 | 47.9 | |
| LLaVA-OV (7B)Tuning Strategy=Zero-Shot, Parameters=7B2026.05 | 2.4 | 44.2 |