fix evaluate

oai-knee-detection/dataset_generation/preprocessing.py

@@ -2,56 +2,52 @@
 import pandas as pd
 import numpy as np
 import os
-import time
 import sys
-import h5py
 from tqdm import tqdm
-import scipy.ndimage as ndimage
+import argparse
 
 '''
 This file used new pipeline to get images preprocessed.
 '''
-MONTH = str(sys.argv[1])
-df = pd.read_csv('../output_data/output{}.csv'.format(MONTH))
-output_folder = '/gpfs/data/denizlab/Users/bz1030/KneeNet/KneeProject/test/bbox_prepprocessing'
-save_dir ='/gpfs/data/denizlab/Users/bz1030/data/OAI_processed_new3'
-if not os.path.exists(output_folder):
-    os.makedirs(output_folder)
-bar = tqdm(total=df.shape[0], desc='Processing', ncols=90)
+parser = argparse.ArgumentParser(description='Annotation')
+parser.add_argument('-m', '--month', action="store", dest="month", type=str,
+                    help='month file for OAI e.g. 00m 12m 24m etc.', default=None)
+parser.add_argument('-md', '--model-dir', type=str, dest='model_dir',
+                    default='../model_weights/KneeJointLocalModel/resnet18_detector.pth')
+parser.add_argument('-sd', '--save-dir', action="store", dest="save_dir", type=str,
+                    help="where you want to save after conversion from DICOM to H5",
+                    default='../data/OAI_processed/')
+def main():
+    args = parser.parse_args()
+    Month = args.month
+    df = pd.read_csv('../output_data/output{}.csv'.format(MONTH))
+    output_folder = '../data/bbox_prepprocessing'
+    save_dir ='../data/OAI_processed_new3'
+    if not os.path.exists(output_folder):
+        os.makedirs(output_folder)
+    bar = tqdm(total=df.shape[0], desc='Processing', ncols=90)
 
-for idx, row in df.iterrows():
-    row = row.tolist()
-    data_path = row[-1]
-    data_path = data_path.replace('//','/')
-    month = data_path.split('/')[-6]
-    p_id = data_path.split('/')[-4]
-    print(month,p_id)
-    bbox = np.array(row[:8])
-    img, data, img_before = image_preprocessing(data_path)
-    left,right = getKneeWithBbox(img,bbox)
-    f_name_l = '{}_{}_LEFT_KNEE.hdf5'.format(p_id,month)
-    f_name_r = '{}_{}_RIGHT_KNEE.hdf5'.format(p_id,month)
-    print(f_name_l,f_name_r)
-    path2save = os.path.join(save_dir,month)
-    print(left.shape,right.shape)
-    left, _, _ = padding(left, img_size=(1024,1024))
-    right, _, _ = padding(right, img_size=(1024,1024))
-    create_h5(path2save,f_name_l,left)
-    create_h5(path2save, f_name_r, right)
-    bar.update(1)
+    for idx, row in df.iterrows():
+        row = row.tolist()
+        data_path = row[-1]
+        data_path = data_path.replace('//','/')
+        month = data_path.split('/')[-6]
+        p_id = data_path.split('/')[-4]
+        print(month,p_id)
+        bbox = np.array(row[:8])
+        img, data, img_before = image_preprocessing(data_path)
+        left,right = getKneeWithBbox(img,bbox)
+        f_name_l = '{}_{}_LEFT_KNEE.hdf5'.format(p_id,month)
+        f_name_r = '{}_{}_RIGHT_KNEE.hdf5'.format(p_id,month)
+        print(f_name_l,f_name_r)
+        path2save = os.path.join(save_dir,month)
+        print(left.shape,right.shape)
+        left, _, _ = padding(left, img_size=(1024,1024))
+        right, _, _ = padding(right, img_size=(1024,1024))
+        create_h5(path2save,f_name_l,left)
+        create_h5(path2save, f_name_r, right)
+        bar.update(1)
 
 
-''' to display the data
-
-for idx, row in df.iterrows():
-    row = row.tolist()
-    data_path = row[-1]
-    data_path = data_path.replace('//','/')
-    bbox = np.array(row[:8])
-    img, data, img_before = image_preprocessing(data_path)
-    f_name = data_path
-    f_name = f_name.split('/')[-6:-1]
-    f_name = '_'.join(f_name) + '.png'
-    drawKneeWithBbox(img,bbox,output_folder,f_name)
-    print(data_path)
-'''
+if __name__ == '__main__':
+    main()

oai-knee-detection/main.py

@@ -19,8 +19,9 @@
                     default='../model_weights/ResNet18_/epoch_45.pth')
 parser.add_argument('-cd', '--content-dir', action="store", dest="content_dir", type=str,
                     help="content file", default='../data/OAI_summary.csv')
+# place the original OAI dataset (DICOM  file) in the data home directory
 parser.add_argument('-dh', '--data-home', action="store", dest="data_home", type=str,
-                    help="data home directory", default='/gpfs/data/denizlab/Datasets/OAI_original')
+                    help="data home directory", default='../data/OAI_original')
 parser.add_argument('-m', '--month', action="store", dest="month", type=str,
                     help="which part of OAI to be processed", default='00m')
 parser.add_argument('-nw', '--num-workers', action="store", dest="num_workers", type=int,

oai-knee-detection/train_test_build.py

@@ -94,9 +94,7 @@ def main(args):
     print(df.shape)
 
     month = '00m'
-    #OAI_DATASET = '/gpfs/data/denizlab/Datasets/OAI_original'
     OAI_DATASET = args.dataset
-    #output_folder = '/gpfs/data/denizlab/Users/bz1030/data/bounding_box/'
     output_folder = args.output_dir
     print(df.shape)
     train = df.sample(frac=0.7)

oai-xray-klg/README.md

@@ -15,11 +15,16 @@ To reproduce the baseline ResNet experiment and ResNet with CBAM, please run the
 ```bash
 python3 main.py -n baseline -m CBAM -do\
  -au -ep 30 -lr 0.0001 -bs 6 -dm no\
- -d /gpfs/data/denizlab/Users/bz1030/data/OAI_processed_new4/\
- -dc /gpfs/data/denizlab/Users/bz1030/data/OAI_proj15/
+ -d ../data/OAI_processed/\
+ -dc ../data/classifier/
 ```
-Above code will get content file from `OAI_proj15` file and then load data from `OAI_processed_new4`. Arguments `-m` can have `baseline` or `CBAM` which stand for the experiments mentioned in paper. Please refer `./main.py` for description of other arguments.
+Above code will get content file from `../data/classifier/` folder and then load data from `OAI_processed` obtained from detector step.
+ Arguments `-m` can have `baseline` or `CBAM` which stand for the experiments mentioned in paper. Please refer `./main.py` for description of other arguments.
 
+To evaluate certain model after train and model selection by best performance on validation dataset. You can run:
+```bash
+python3 evaluate.py -lm ../model_weights/Classifier/best-model-baseline.pth -hd ../data/classifier/
+```
 To reproduce all attention heatmap, please run the following code
 ```bash
 python3 attention_map.py -lm <model-weights> -hd <home-dir-where-you-saved-h5-files> -sp <summary-path>