Clock-time recognition on Clock-Time
97.56AccuracySMMD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SMMDModel=Ministral-3, Params=3B, Loss=Ours2026.06 | 97.56 | 8.07 | |
| GCEModel=Ministral-3, Params=3B, Loss=GCE2026.06 | 93.82 | 10.39 | |
| NTILModel=Ministral-3, Params=3B, Loss=NTIL2026.06 | 90.21 | 23.88 | |
| NTLModel=Ministral-3, Params=3B, Loss=NTL2026.06 | 89.17 | 25.48 | |
| CEModel=Ministral-3, Params=3B, Loss=CE2026.06 | 82.36 | 37.36 | |
| GCEModel=Llama-3.2-Vision, Params=11B, Loss=GCE2026.06 | 82.15 | 28.63 | |
| SMMDModel=Llama-3.2-Vision, Params=11B, Loss=Ours2026.06 | 80.76 | 32.45 | |
| NTLModel=Llama-3.2-Vision, Params=11B, Loss=NTL2026.06 | 79.79 | 33.29 | |
| SMMDModel=Qwen2.5-VL, Params=7B, Loss=Ours2026.06 | 78.26 | 41.07 | |
| NTLModel=Qwen2.5-VL, Params=7B, Loss=NTL2026.06 | 75.97 | 41.45 | |
| NTILModel=Qwen2.5-VL, Params=7B, Loss=NTIL2026.06 | 75.28 | 45.74 | |
| SMMDModel=Qwen2.5-VL, Params=3B, Loss=Ours2026.06 | 74.3 | 50.59 | |
| CEModel=Llama-3.2-Vision, Params=11B, Loss=CE2026.06 | 71.8 | 46.64 | |
| NTILModel=Llama-3.2-Vision, Params=11B, Loss=NTIL2026.06 | 69.51 | 55.09 | |
| NTILModel=Qwen2.5-VL, Params=3B, Loss=NTIL2026.06 | 69.46 | 55.43 | |
| GCEModel=Qwen2.5-VL, Params=7B, Loss=GCE2026.06 | 69.23 | 42.95 | |
| NTLModel=Qwen2.5-VL, Params=3B, Loss=NTL2026.06 | 68.68 | 57.73 | |
| CEModel=Qwen2.5-VL, Params=7B, Loss=CE2026.06 | 65.69 | 59.11 | |
| CEModel=Qwen2.5-VL, Params=3B, Loss=CE2026.06 | 46.39 | 86.73 | |
| GCEModel=Qwen2.5-VL, Params=3B, Loss=GCE2026.06 | 38.4 | 55.28 |