Tongue Report Generation on MedTCM 1.0 (test)
83.57BLEU-4MMIR–TCM (Mask train + Mask)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MMIR–TCM (Mask train + Mask)Backbone=Qwen3–VL 30B, Params=30B, LoRA r=64, Epochs=10, Input Type=Masked images2026.07 | 83.57 | 77.3 | 76.77 | 84.3 | 62.7 | 47.3 | 69.3 | 64.9 | |
| MMIR–TCM (Mask train + Raw)Backbone=Qwen3–VL 30B, Params=30B, LoRA r=64, Epochs=10, Input Type=Masked training, Raw inference2026.07 | 83.22 | 76.06 | 75.69 | 83.45 | 61.7 | 47.8 | 69.3 | 59.2 | |
| MMIR–TCM (Raw train + Raw)Backbone=Qwen3–VL 30B, Params=30B, LoRA r=64, Epochs=10, Input Type=Raw images2026.07 | 83.2 | 76.24 | 76.06 | 83.62 | 60.1 | 44.2 | 67.3 | 61.1 | |
| MMIR–TCM (Raw train + Mask)Backbone=Qwen3–VL 30B, Params=30B, LoRA r=64, Epochs=10, Input Type=Raw training, Masked inference2026.07 | 83.11 | 75.96 | 75.57 | 83.34 | 61.2 | 47.2 | 68.3 | 59 | |
| MMIR–TCM (Mask train + Mask, 3ep)Backbone=Qwen3–VL 30B, Params=30B, LoRA r=64, Epochs=3, Input Type=Masked images2026.07 | 83.09 | 76.5 | 76.28 | 83.73 | 61.1 | 43.9 | 69.8 | 63.2 | |
| MMIR–TCM (Raw train + Raw, 3ep)Backbone=Qwen3–VL 30B, Params=30B, LoRA r=64, Epochs=3, Input Type=Raw images2026.07 | 81.72 | 74.42 | 73.88 | 82.2 | 58 | 38.5 | 68.2 | 60.7 | |
| ViTCM–LLM E4 Full (r=64, 20ep)Backbone=Qwen2.5–VL 32B, Params=32B, LoRA r=64, Epochs=202026.07 | 44.07 | — | — | — | 61.5 | — | — | — | |
| ViTCM–LLM E4 Full (r=64, 10ep)Backbone=Qwen2.5–VL 32B, Params=32B, LoRA r=64, Epochs=102026.07 | 43.57 | — | — | — | 58.58 | — | — | — | |
| ViTCM–LLM E4 Full (r=64, 3ep)Backbone=Qwen2.5–VL 32B, Params=32B, LoRA r=64, Epochs=32026.07 | 39.57 | — | — | — | 39.15 | — | — | — | |
| ViTCM–LLM E4 Full (r=16, 3ep)Backbone=Qwen2.5–VL 32B, Params=32B, LoRA r=16, Epochs=32026.07 | 37.94 | — | — | — | 36.48 | — | — | — | |
| ViTCM–LLM E3 Vision+ProjectorBackbone=Qwen2.5–VL 32B, Params=32B, LoRA r=16, Epochs=32026.07 | 35.82 | — | — | — | 36.1 | — | — | — | |
| Original Qwen3–VL 30B (untuned)Params=30B2026.07 | 33.98 | 28.42 | 28.23 | 34.05 | 28.9 | 18.7 | 28.8 | 14.1 | |
| ViTCM–LLM E2 Language–onlyBackbone=Qwen2.5–VL 32B, Params=32B, LoRA r=16, Epochs=32026.07 | 26.74 | — | — | — | 26.98 | — | — | — | |
| Grok–2–Vision–1212Setting=zero-shot2026.07 | 26.4 | 22.12 | 22.08 | 26.43 | 33.8 | 26.2 | 40.4 | 36.7 | |
| LLaMA4–scout 109BParams=109B, Setting=zero-shot2026.07 | 25.68 | 22.45 | 22.43 | 25.7 | 33.6 | 25.4 | 41.2 | 35.6 | |
| Gemini–2.5–FlashSetting=zero-shot2026.07 | 24.43 | 21.37 | 21.33 | 24.46 | 35.3 | 27.1 | 42.1 | 37.2 | |
| ViTCM–LLM E1 Zero–shotBackbone=Qwen2.5–VL 32B, Params=32B2026.07 | 24.08 | — | — | — | 35.38 | — | — | — |