From 9fe688d81ed9877a6fbe140267f7b627a743a075 Mon Sep 17 00:00:00 2001
From: Umar Khan <umee909@gmail.com>
Date: Thu, 16 Apr 2020 19:23:23 +0530
Subject: [PATCH] Convert data set to CSV with custom resize

This code creates a CSV format of data set with resized images of your defined size (Height, Width)
We got labels for every image from metadata and in new created dataset, first attribute defines the label
label == 1 defines COVID-19 and Label == 0 defines all others
Since Excel contains XFD as last column and XFD equals 16384. Thus we have to look for the resized height and width
In this example we took 40 x 40 with 3 channels making total row size (40x40x3)+1 +1 is added because label is stored as first attribute of image
---
 GetCovidToCSV.py | 85 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 85 insertions(+)
 create mode 100644 GetCovidToCSV.py

diff --git a/GetCovidToCSV.py b/GetCovidToCSV.py
new file mode 100644
index 00000000..c405f1da
--- /dev/null
+++ b/GetCovidToCSV.py
@@ -0,0 +1,85 @@
+'''
+This code creates a CSV format of data set with resized images of your defined size (Height, Width)
+We got labels for every image from metadata.csv and in new creted dataset, first attribute defines the label
+label == 1 defines COVID-19 and Label == 0 defines all others
+Since Excel contains XFD as last column and XFD equals 16384. Thus we have to look for the resized height and width
+In this example we took 40 x 40 with 3 channels making total row size (40x40x3)+1 +1 is added because label is stored as first attribute of image
+
+'''
+import numpy as np
+import pandas as pd
+import csv
+from keras.preprocessing.image import ImageDataGenerator
+import matplotlib.pyplot as plt
+
+
+#Defining image size
+Height = 40
+Width = 40
+
+
+#defining dataset dir
+DATASET_DIR = "./dataset/Images/" #Define paths as per your environment
+metadata = "./dataset/metadata/metadata.csv" # Meta info - path to metadata file
+
+
+# Resizing Training data images to 40 x 40
+train_datagen = ImageDataGenerator()
+train_generator = train_datagen.flow_from_directory(
+    DATASET_DIR,
+    target_size=(Height, Width),
+    batch_size=1,
+    class_mode='binary',
+    subset='training')
+
+
+#Getting Labels
+metadata_csv = pd.read_csv(metadata)
+labels = []
+for file in train_generator.filenames:
+    for (x, row) in metadata_csv.iterrows():
+        file_name = 'images\\'+ row["filename"] #file contains the directory path 'images\' as well. Thus prefixing images\ to row['filename'] for comparison
+        if file_name == file:
+            label = row['finding']
+            if label == "COVID-19":
+                label = 1 #label == 1 defines COVID-19
+            else:
+                label = 0
+            break
+    labels.append(label)
+
+print('Total Images: ',len(labels))
+
+
+#Writting pixel values to CSV file
+with open('mydata.csv', 'w', newline='') as file:
+    writer = csv.writer(file)
+    for i in range(len(train_generator)):
+        image, label = train_generator[i]
+        image = image.reshape(Height*Width*3) # Reshaping into single list
+        label = labels[i]     
+        row = [label] #Writting label as first element of the row 
+        for i in range(len(image)): #Iterating over Height
+            row.append(image[i]) #Adding all pixels of an image in a single row after the label
+        writer.writerow(row)
+
+print("CSV data genereated sucessfully !");
+
+
+# Plot the first image from the input dataset to be sure that data is correctly transformed
+#============================================================
+# Take a single image, and remove the color dimension by reshaping
+imgSize = Height*Width*3 + 1 #We used first element as label
+image = row[1:imgSize] #Taking Image pixels from element 2
+image = np.array(image)
+image = image.reshape(Height,Width,3)
+
+
+#Plotting here
+plt.figure()
+plt.imshow(image.astype(int), cmap=plt.cm.binary)
+plt.colorbar()
+plt.grid(False)
+plt.title('First Image')
+plt.show()
+