Semantic Segmentation on AOI single device layout
71.5mIoUPatch Retrieval
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Patch RetrievalStrategy=Patch Retrieval2026.05 | 71.5 | 68 | 65.6 | 78.7 | 73.5 | |
| FasterViT-0 + UPerNetPre-training=AOI-MAE, Throughput (Crops/s)=163.32026.05 | 60.3 | 79.3 | 66.7 | 19.3 | 75.8 | |
| ViT-Tiny + UPerNetPre-training=AOI-MAE, Throughput (Crops/s)=218.12026.05 | 53.5 | 73.7 | 43 | 26.5 | 70.7 | |
| ResNet18 + U-Net++Training=Supervised Baseline, Throughput (Crops/s)=86.92026.05 | 52.4 | 75.4 | 59.1 | 9.2 | 65.8 | |
| ViT-Tiny + Patch RetrievalPre-training=AOI-MAE, Throughput (Crops/s)=266.72026.05 | 48.1 | 38.4 | 47.8 | 44.4 | 61.8 | |
| ViT-Small Patch RetrievalPre-training=Frozen DINOv2, Throughput (Crops/s)=90.92026.05 | 47.8 | 43.4 | 48.7 | 45.8 | 53.4 | |
| ResNet18 + U-Net++Strategy=Fine-tuned2026.05 | 47.5 | 43.3 | 72.1 | 3.4 | 71.1 | |
| ResNet18 + DeepLabTraining=Supervised Baseline, Throughput (Crops/s)=87.32026.05 | 43.5 | 66.7 | 50.3 | 10.3 | 46.7 |