Letter Recognition on Hell-Char (20% stratified test)
83.1AccuracyResNet18-PT+FT (ours)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ResNet18-PT+FT (ours)Model=ResNet18-PT+FT, Fragmentation=LF, Contrastive Loss=DSCL2026.06 | 83.1 | 83 | |
| ResNet18-PT+FTModel=ResNet18-PT+FT, Fragmentation=none, Contrastive Loss=SCL2026.06 | 81.8 | 81 | |
| fCNN (ours)Model=fCNN, Fragmentation=LF, Contrastive Loss=DSCL2026.06 | 80.3 | 80 | |
| ResNet18-PT+FTModel=ResNet18-PT+FT, Fragmentation=none, Contrastive Loss=none2026.06 | 80.1 | 79 | |
| ResNet18-FTModel=ResNet18-FT, Fragmentation=none, Contrastive Loss=none2026.06 | 78.8 | 74 | |
| fCNNModel=fCNN, Fragmentation=LF, Contrastive Loss=none2026.06 | 78.2 | 77 | |
| fCNNModel=fCNN, Fragmentation=RE, Contrastive Loss=none2026.06 | 76.8 | 75 | |
| fCNNModel=fCNN, Fragmentation=none, Contrastive Loss=none2026.06 | 74.2 | 74 |