updated and cleaned repo, added readme

Author: architkaila

README.md

@@ -1,18 +1,8 @@
-Flask
-matplotlib
-monai
-networkx
-opencv_python
-pandas
-Pillow
-pydicom
-requests
-scikit_learn
-seaborn
-streamlit
-torch
-torchsummary
-torchtext
-torchvision
-tqdm
-numpy
+pandas==1.5.3
+numpy==1.23.5
+opendatasets==0.1.22
+Pillow==9.4.0
+torch==2.0.0
+torchsummary==1.5.1
+torchvision==0.15.1
+tqdm==4.65.0

assets/network_architecture.png

@@ -0,0 +1 @@
+## Placeholder to make this directory a package

assets/sample_hr_input.png

@@ -15,5 +15,5 @@ def download_dataset(dataset_url):
 if __name__ == "__main__":
 
     ## Download the dataset
-    dataset_url = 'https://www.kaggle.com/andrewmvd/heart-failure-clinical-data'
+    dataset_url = 'https://www.kaggle.com/datasets/nih-chest-xrays/data'
     download_dataset(dataset_url)

assets/sample_lr_input.png

@@ -1,3 +1,6 @@
+## SSIM implementation sourced from the original author's GitHub repository of the SRGAN paper
+# https://github.com/goldhuang/SRGAN-PyTorch/blob/master/pytorch_ssim/__init__.py
+
 ## Library Imports
 import math
 import torch

assets/sample_sr_output.png

@@ -15,7 +15,7 @@ def split_data():
         None
     """
     ## Get list of all images in full dataset
-    all_images_list = glob.glob(f"./dataset/*/*/*.png", recursive=False)
+    all_images_list = glob.glob(f"./data/*/*/*.png", recursive=False)
 
     ## Shuffle the data
     random.shuffle(all_images_list)
@@ -25,9 +25,9 @@ def split_data():
     test_images = all_images_list[90000:]
 
     ## Save the train and test split
-    with open("./dataset/train_images.pkl", "wb") as fp:
+    with open("./data/train_images.pkl", "wb") as fp:
         pickle.dump(train_images, fp)
-    with open("./dataset/val_images.pkl", "wb") as fp:
+    with open("./data/val_images.pkl", "wb") as fp:
         pickle.dump(test_images, fp)
 
 

dataset/train_images.pkl renamed to data/train_images.pkl

@@ -31,7 +31,7 @@ def run_pipeline(arguments):
 
     ## Create directories to store results if they don't exist
     print("[INFO] Creating directories")
-    os.makedirs("./checkpoints", exist_ok=True)
+    os.makedirs("./models", exist_ok=True)
     os.makedirs("./logs", exist_ok=True)
 
     ## Initialize parameters from arguments
@@ -41,9 +41,9 @@ def run_pipeline(arguments):
     DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
     ## Load Train and Val data splits
-    with open('./dataset/train_images.pkl', 'rb') as f:
+    with open('./data/train_images.pkl', 'rb') as f:
         train_data_list = pickle.load(f)
-    with open('./dataset/val_images.pkl', 'rb') as f:
+    with open('./data/val_images.pkl', 'rb') as f:
         val_data_list = pickle.load(f)
 
     ## Load the train dataset
@@ -276,7 +276,7 @@ def run_pipeline(arguments):
 
         ## Save the results to the disk
         data_frame.to_csv(
-            out_path + "ssrgan_" + str(epoch) + "_train_results.csv",
+            out_path + "metrics_" + str(epoch) + "_train_results.csv",
             index_label="Epoch",
         )