cryo-sam 0.0.73

A tool to find cryo-em targets for cyro-em automation building on foundation ML models.

Cryo-SAM Author: James Rodgers, Aditya Balu Organization: Iowa State University - Department of Electrical and Computer Engineering Email: [e-mail]

Uses Sam for the detection of squares, holes, and particles in cryo-em.

Usage:

from cryo_sam import Csam

Model checkpoint location, several types downloadable from https://github.com/facebookresearch/segment-anything#model-checkpoints testing done with the below model and checkpoint information:

model_name = "vit_h"
checkpoint = "sam_model/sam_vit_h_4b8939.pth" 
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

Example construction: detector = Csam(model_name, checkpoint, device)

Csam.hole_finder(numpy_image_array) Parameter: numpy image array Returns: list of square coordinates (x, y, w, h) Csam.square_finder(numpy_image_array): Parameter: numpy image array Returns: list of square coordinates (x, y, w, h)

Both of the above returns will need to be changed in the future since extra data is to be returned be the exact nature of that isn't determined yet. It will probably return a class containing the data in the near future.

Csam.visualize_detections(image, image_scale_percent, box_prompts = None, point_prompts = None, show_boxes = True, show_point_prompts = None, show_box_prompts, show_masks = True): Parameters: image: numpy image array, numpy array of points, numpy array of bounding box, bool, bool, bool image_scale_percent: int representing scaled percent i.e 60 = 60% of original size point_prompts: numpy array of points. [[x, y], [x, y], [x, y], ...] box_prompts: numpy array of rectangle boxes [[x1, y1, x2, y2], [x1, y1, x2, y2], ...] show_boxes: bool, show bounding boxes generated by the model show_masks: bool, show masks generated by the model show_point_prompts: bool, show point prompts sent to the model show_box_prompts: bool, show bounding box prompts sent to the model Returns: None

Example usage:

test_image = np.load("/mnt/big_data_2/pncc/annotated_data/160002/box4_g11/data_031022/mmm_0.npy")
test_image = cv2.cvtColor(test_image, cv2.COLOR_BGR2RGB)

print(detector.hole_finder(test_image))
detector.visualize_detections(test_image, image_scale_percent, show_boxes=True)

image_scale_percent = 40

This function will display the image with the detections overlaid on top of it. It is currently not specialized to hole or square.
This will be changed in the future, this is mostly for testing purposes.
The image_scale_percent is the percent of the original image size that the image will be displayed at. 
This is useful for large images that would be too large to display on a screen and to improve render speed. 
The default is 50% of the original size. Will probably add a save image option in the future.