Table 6 Computational complexity comparison of different methods.
From: A hybrid YOLO-UNet3D framework for automated protein particle annotation in Cryo-ET images
Method | Parameters (M) | GPU Memory (GB) | Inference Time (s) | Processing Speed (voxels/s) |
|---|---|---|---|---|
DeepFinder | 8.5 | 3.2 | 42.6 | \(1.6 \times 10^6\) |
UNet3D | 16.8 | 5.6 | 68.3 | \(1.0 \times 10^6\) |
YOLOv5-3D | 13.5 | 4.8 | 34.2 | \(2.0 \times 10^6\) |
UNet3D + DBSCAN | 16.8 | 5.7 | 75.5 | \(0.9 \times 10 ^6\) |
YOLOv5-3D + DBSCAN | 13.5 | 4.9 | 38.1 | \(1.8 \times 10^6\) |
Ours | 24.8 | 7.8 | 56.8 | \(1.2 \times 10^6\) |
