Web Image Translation on Multilingual Web Image Translation HI-EN
41.3BLEUVaaWIT on Qwen3 (8B)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VaaWIT on Qwen3 (8B)Backbone=Qwen3 (8B), Tuning Strategy=DSAM + VAA2026.05 | 41.3 | 85 | |
| VaaWIT on LLaMA3.1 (8B)Backbone=LLaMA3.1 (8B), Tuning Strategy=DSAM + VAA2026.05 | 35.7 | 83.9 | |
| Full Fine-TuningBackbone=Qwen3-VL (8B), Tuning Strategy=Full Fine-Tuning2026.05 | 34.7 | 73 | |
| EasyOCR_Google APIType=Cascaded Model2026.05 | 21.4 | 65.3 | |
| Gemini2.5 Pro†Tuning Strategy=Zero-Shot2026.05 | 17.5 | 70.9 | |
| GPT4.1†Tuning Strategy=Zero-Shot2026.05 | 17.3 | 69.9 | |
| PP-OCR_Microsoft APIType=Cascaded Model2026.05 | 15.4 | 67.4 | |
| Chain-of-ThoughtBackbone=Qwen3-VL (8B), Tuning Strategy=Prompting2026.05 | 14.7 | 69 | |
| LoRABackbone=Qwen3-VL (8B), Tuning Strategy=LoRA (r=8)2026.05 | 11.1 | 65.2 | |
| Qwen3-VL (32B)Tuning Strategy=Zero-Shot, Parameters=32B2026.05 | 10.5 | 63.3 | |
| Qwen3-VL (8B)Tuning Strategy=Zero-Shot, Parameters=8B2026.05 | 10.1 | 63.1 | |
| LLaMA3.2 (90B)Tuning Strategy=Zero-Shot, Parameters=90B2026.05 | 2.8 | 49.2 | |
| LLaMA3.2 (11B)Tuning Strategy=Zero-Shot, Parameters=11B2026.05 | 2.1 | 45.8 | |
| LLaVA-OV (7B)Tuning Strategy=Zero-Shot, Parameters=7B2026.05 | 1.6 | 44.2 |