Table 2 Parameters used for fine tuning of the Segment Anything Model (SAM), along with their corresponding descriptions and implications
From: Data-driven microstructural optimization of Ag-Bi-I perovskite-inspired materials
Parameter | Description | Implication |
|---|---|---|
points_per_side | Number of sampling points along each side of the image grid. | Higher values improve granularity but increase computation. |
pred_iou_thresh | Minimum IoU threshold for mask predictions. | Stricter thresholds (e.g., 0.7–0.9) enhance precision but may exclude valid masks. |
stability_score_thresh | Filters masks based on their stability across scales. | Higher values (e.g., 0.9–0.95) favor stable masks but risk discarding borderline regions. |
crop_n_layers | Number of hierarchical cropping layers for segmentation. | More layers improve segmentation of small objects but increase processing time. |
crop_n_points_downscale_factor | Downscales the number of points sampled in cropped regions. | Smaller factors preserve resolution but increase computational cost. |
min_mask_region_area | Minimum size of mask regions | Filters small noisy masks. Set this based on object size in your dataset (e.g., around 0.1–1% of image resolution). |
