Defect Classification on SemiFA-930 (val)
90AccuracyDINOv2 + MLP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DINOv2 + MLPTrainable Params=214K2026.04 | 90 | 89.8 | |
| ResNet-50Evaluation Protocol=e2e fine-tuned, Trainable Params=23.5M2026.04 | 82.9 | 83.9 | |
| CLIP ViT-B/32Evaluation Protocol=zero-shot, Trainable Params=02026.04 | 20.7 | 17.3 |