Table 4 Comparison of system usage and complexity metrics between DINOv2 and HistoDARE during feature extraction.
From: A histopathology aware DINO model with attention based representation enhancement
Metric | DINOv2 | HistoDARE | Change |
|---|---|---|---|
Model FLOPs (MACs) | 77.89 GMac | 78.42 GMac | +0.53 GMac |
Number of Parameters | 304.37 M | 306.60 M | +2.23 M |
Feature Extraction Time (s) | 2080.08 | 2097.84 | +17.76 s |
System Usage Before Feature Extraction | |||
CPU RAM Usage | 1.71 GB (3.80%) | 21.34 GB (35.40%) | +19.63 GB |
GPU VRAM Usage | 1911 MB (11.66%) | 15542 MB (94.86%) | +13631 MB |
Max CUDA Allocated (PyTorch) | 1753.22 MB | 8257.58 MB | +6504.36 MB |
System Usage After Feature Extraction | |||
CPU RAM Usage | 2.42 GB (5.00%) | 21.58 GB (35.80%) | +19.16 GB |
GPU VRAM Usage | 2471 MB (15.08%) | 15542 MB (94.86%) | +13071 MB |
Max CUDA Allocated (PyTorch) | 1640.45 MB | 8058.91 MB | +6418.46 MB |
