Pneumothorax_prediction

import import_ipynb
from utils import mask2rle,combined_loss,dice_coef
import cv2
import tensorflow as tf
from tensorflow import keras
from tqdm import tqdm
def predict_results(test_files_png):
    '''
         Given a list of chest Xray image files it generates prediction results for Pneumothorax disease
        Args:
            test_files_png : List of files for which we need to generate predictions
        Returns:
                None
    '''
    model_seg = tf.keras.models.load_model('models/Uefficientnetb4',custom_objects={'combined_loss' :combined_loss, \
                                                                             'dice_coef' : dice_coef})
    d = dict()
    for file in tqdm(test_files_png):
        img = tf.io.read_file(file)
        img = tf.image.decode_png(img, channels= N_CHANNELS)
        img = tf.image.convert_image_dtype(img, tf.float32) 
        img = tf.image.resize(img, [IMG_WIDTH, IMG_HEIGHT]) 
        img.set_shape((IMG_HEIGHT,IMG_WIDTH,3))
        pred_mask= model_seg.predict(tf.expand_dims(img,axis=0)).reshape((IMG_HEIGHT,IMG_WIDTH))
        if (pred_mask > .5).astype(int).sum() >0 :
            pred_mask = cv2.resize(pred_mask,(1024,1024))
            pred_mask = (pred_mask > .5).astype(int)
            #plt.imshow(pred_mask.squeeze())
            d[os.path.splitext(file.split('/')[-1])[0]] = mask2rle(pred_mask.T * 255, 1024,1024)
        else:
            d[os.path.splitext(file.split('/')[-1])[0]] = '-1'
    
    #Convert the dictionary into a DataFrame
    sub = pd.DataFrame.from_dict(d, orient='index')
    #Reset index
    sub.reset_index(inplace=True)
    #Set column names
    sub.columns = ['ImageId', 'EncodedPixels']
    sub.head()
    sub.to_csv('submission.csv', index=False, header=True)