Table 2 Inference model performance comparison.
Methods | Visual genome | MIMIC-CXR | nuScenes | ||||
|---|---|---|---|---|---|---|---|
Acc(%) | F1(%) | CB-F1 score (%) | AUC(%) | F1(%) | mAP(%) | F1(%) | |
MetaMath | 71.2 | 68.5 | 65.0 | 85.0 | 72.1 | 67.5 | 70.3 |
CausalBERT | 75.6 | 72.3 | 70.2 | 88.0 | 75.8 | 71.4 | 74.5 |
QRNN | 78.9 | 75.8 | 73.5 | 90.0 | 78.2 | 74.2 | 77.6 |
DeepSeek-R1:7B | 83.1 | 80.2 | 76.8 | 92.3 | 82.5 | 79.5 | 80.6 |
CLIP-ViL | 81.7 | 78.5 | 75.2 | 91.5 | 81.3 | 78.1 | 79.4 |
UNITER | 80.9 | 77.6 | 74.7 | 90.7 | 80.5 | 77.9 | 78.6 |
CDMRNet | 89.7 | 84.1 | 82.0 | 96.0 | 85.4 | 83.1 | 83.9 |