generate_starfile_new_data_jpg.py

# Code for generating star file

from utils.denoise import denoise, denoise_jpg_image
import config
import matplotlib.pyplot as plt
import numpy as np
import torch
import cv2
import csv
import glob
import random
from dataset.dataset import transform, min_max
from models.model_5_layers import UNET
import config
from tqdm import tqdm
import mrcfile
from segment_anything import SamAutomaticMaskGenerator, sam_model_registry
import statistics as st
import config

print("[INFO] Loading up model...")
model = UNET().to(device=config.device)
state_dict = torch.load(config.cryosegnet_checkpoint)
model.load_state_dict(state_dict)

sam_model = sam_model_registry[config.model_type](checkpoint=config.sam_checkpoint)
sam_model.to(device=config.device)

mask_generator = SamAutomaticMaskGenerator(sam_model)

def get_annotations(anns):
    if len(anns) == 0:
        return
    sorted_anns = sorted(anns, key=(lambda x: x['area']), reverse=True)
    ax = plt.gca()
    ax.set_autoscale_on(False)

    img = np.ones((sorted_anns[0]['segmentation'].shape[0], sorted_anns[0]['segmentation'].shape[1], 4))
    img[:,:,3] = 0
    for ann in sorted_anns:
        m = ann['segmentation']
        color_mask = np.concatenate([np.random.random(3), [0.35]])
        img[m] = color_mask
    
    return img

def generate_output(model, image_path, star_writer):
    # set model to evaluation mode
    model.eval()
    # turn off gradient tracking
    with torch.no_grad():        
        image = cv2.imread(image_path, 0)
        
        #Check if denoising makes difference or not! If the images are already denoised don't denoise them else denoise them!
        image = denoise_jpg_image(image)
        height, width = image.shape
        image = cv2.resize(image, (config.input_image_width, config.input_image_height))
        
        image = torch.from_numpy(image).unsqueeze(0).float()
        image = image / 255.0
        image = image.to(config.device).unsqueeze(0)
        
        predicted_mask = model(image)       
        predicted_mask = torch.sigmoid(predicted_mask)
        predicted_mask = predicted_mask.cpu().numpy().reshape(config.input_image_width, config.input_image_height)
        
        # Uncomment these two lines if you find positions of predicted proteins are flipped
        #predicted_mask = np.rot90(predicted_mask, k=3)
        #predicted_mask = predicted_mask.T
                
        sam_output = np.repeat(transform(predicted_mask)[:,:,None], 3, axis=-1)
        predicted_mask = cv2.resize(predicted_mask, (width, height))
        predicted_mask = min_max(predicted_mask) 
        
        masks = mask_generator.generate(sam_output)
        sam_mask = get_annotations(masks)
        sam_mask = cv2.resize(sam_mask, (width, height) )
        
        
        bboxes = {"bbox": [], "iou": []}
        for i in range(0, len(masks)):
            if masks[i]["predicted_iou"] > 0.94:
                bboxes["bbox"].append(masks[i]["bbox"])
                bboxes["iou"].append(masks[i]["predicted_iou"])
        
        if len(bboxes) > 1:
            x_ = st.mode([box[2] for box in bboxes["bbox"]])
            y_ = st.mode([box[3] for box in bboxes["bbox"]])
            d_ = np.sqrt((x_ * width / config.input_image_width)**2 + (y_ * height / config.input_image_height)**2)
            r_ = int(d_//2)
            th = r_ * 0.2     

            filename = image_path.split("/")[-1][:-4] + '.mrc'
            for i in range(len(bboxes["bbox"])):
                box, iou = bboxes["bbox"][i], bboxes["iou"][i] 
                if box[2] < x_ + th and box[2] > x_ - th/3 and box[3] < y_ + th and box[3] > y_ - th/3:                 
                    x_new, y_new = int((box[0] + box[2]/2) / config.input_image_width * width) , int((box[1] + box[3]/2) / config.input_image_height * height)
                    star_writer.writerow([filename, x_new, y_new, 2*r_]) 
                    if iou > 0.9999:
                        star_writer.writerow([filename, x_new + random.randint(-int(r_ / 10), int(r_ / 10)), y_new + random.randint(-int(r_ / 10), int(r_ / 10)), 2*r_])   
                                                                                                                                                                    
        else:
            pass
        
print("[INFO] Loading up Test Micrographs ...")
images_path = list(glob.glob(f"{config.my_dataset_path}/*.*p*g"))

print(f"[INFO] Number of Micrographs = {len(images_path)}\n")
print("[INFO] Generating star file for input Cryo-EM Micrographs...")
print("[INFO] Generation may take more time depending upon the number of micrographs...\n")

with open(f"{config.output_path}/star_files/{config.file_name}", "w") as star_file:
    star_writer = csv.writer(star_file, delimiter=' ')
    star_writer.writerow([])
    star_writer.writerow(["data_"])
    star_writer.writerow([])
    star_writer.writerow(["loop_"])
    star_writer.writerow(["_rlnMicrographName", "#1"])
    star_writer.writerow(["_rlnCoordinateX", "#2"])
    star_writer.writerow(["_rlnCoordinateY", "#3"])
    star_writer.writerow(["_rlnDiameter", "#4"])

    for i in tqdm(range(0, len(images_path), 1)):
        generate_output(model, images_path[i], star_writer)