checkpoints updataed, scripts updated, dataset links updated

abzargar · Apr 20, 2023 · d02dc2d · d02dc2d
1 parent 264b718
commit d02dc2d
Show file tree

Hide file tree

Showing 14 changed files with 191 additions and 2,056 deletions.
diff --git a/README.md b/README.md
@@ -11,9 +11,9 @@ To download our datasets go to https://deepseas.org/datasets/ or:
 
 * Link to [Original annotated dataset](https://drive.google.com/drive/folders/13RhhBAetSWkjySyhJcDqj_FaO09hxkhO?usp=sharing)
 
-* Link to [dataset example for cell segmentation](https://drive.google.com/drive/folders/1gJIkwUQEtut4JCCoUXUcKUWp2gVYxQ9P?usp=sharing)
+* Link to [dataset example for cell segmentation](https://drive.google.com/drive/folders/18odgkzafW8stHkzME_s7Es-ue7odVAc5?usp=sharing)
 
-* Link to [dataset example for cell tracking](https://drive.google.com/drive/folders/17n0Ex8NQS-REB5ZAMlntVnYBnSmZJtLR?usp=sharing)
+* Link to [dataset example for cell tracking](https://drive.google.com/drive/folders/10LWey85fgHgFj_myIr1CYSOviD4SleE4?usp=sharing)
 
 ### Pre-trained models
 They are saved in the folder "trained_models".

diff --git a/deepsea/data.py b/deepsea/data.py
@@ -17,7 +17,7 @@ def __init__(self, images_dir: str, masks_dir: str,unetwmaps_dir: str, transform
         if if_train_aug:
             self.ids = [os.path.splitext(file)[0] for file in sorted(os.listdir(images_dir)) if not file.startswith('.')]
             tmp=[]
-            for i in range(train_aug_iter):
+            for i in range(train_aug_iter+1):
                 tmp+=self.ids
             self.ids=tmp
         else:
@@ -42,14 +42,17 @@ def __getitem__(self, idx):
         assert len(mask_file) == 1, f'Either no mask or multiple masks found for the ID {name}: {mask_file}'
         assert len(img_file) == 1, f'Either no image or multiple images found for the ID {name}: {img_file}'
 
-        mask = cv2.imread(mask_file[0].as_posix())[:, :, 0] > 0
-        wmap=cv2.imread(unetwmap_file[0].as_posix())[:, :, 0]
+        mask = cv2.imread(mask_file[0].as_posix(),0) > 0
 
-        mask=mask & (wmap<255)
-        mask = Image.fromarray(np.uint8(mask.astype('float32') * 255))
-        wmap = Image.fromarray(np.uint8(wmap))
+        wmap=cv2.imread(unetwmap_file[0].as_posix(),0)>0
+
+
+        mask=mask & (wmap<1)
+        wmap=wmap.astype('float32')
+        mask = mask.astype('float32')
+        img=cv2.imread(img_file[0].as_posix(),0).astype('float32')
+        img=(255 * ((img - img.min()) / (img.ptp()+1e-6))).astype(np.uint8)
 
-        img = Image.open(img_file[0])
 
 
         assert img.size == mask.size, \
@@ -169,9 +172,6 @@ def preprocess(cls, pil_img_prev, pil_curr,pil_mask,transforms):
 
     def __getitem__(self, idx):
         img_prev,img_curr,mask = self.ids[idx]
-        img_prev=Image.fromarray(img_prev)
-        img_curr = Image.fromarray(img_curr)
-        mask = Image.fromarray(mask)
         tensor_img_prev,tensor_img_curr,tensor_mask = self.preprocess(img_prev,img_curr,mask,self.transforms)
 
         return {

diff --git a/deepsea/evaluate_test_set_segmentation.py b/deepsea/evaluate_test_set_segmentation.py
@@ -1,66 +1,3 @@
-<<<<<<< HEAD:evaluate_test_set_segmentation.py
-import torch.utils.data as data
-import segmentation_transforms as transforms
-import numpy as np
-import argparse
-import os
-import random
-from model import DeepSeaSegmentation
-from data import BasicSegmentationDataset
-import torch
-from evaluate import evaluate_segmentation
-from utils import get_n_params
-
-SEED = 1234
-random.seed(SEED)
-np.random.seed(SEED)
-torch.manual_seed(SEED)
-torch.cuda.manual_seed(SEED)
-torch.backends.cudnn.deterministic = True
-
-
-def test(args,image_size = [383,512],image_means = [0.5],image_stds= [0.5],batch_size=1):
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-
-    test_transforms = transforms.Compose([
-                               transforms.Resize(image_size),
-                               transforms.Grayscale(num_output_channels=1),
-                               transforms.ToTensor(),
-                               transforms.Normalize(mean = image_means,
-                                                    std = image_stds)
-                           ])
-
-
-    test_data = BasicSegmentationDataset(os.path.join(args.test_set_dir, 'images'), os.path.join(args.test_set_dir, 'masks'),os.path.join(args.test_set_dir, 'wmaps'),transforms=test_transforms)
-
-    test_iterator = data.DataLoader(test_data,batch_size = batch_size,shuffle=False)
-
-    model=DeepSeaSegmentation(n_channels=1, n_classes=2, bilinear=True)
-    print('INFO: Num of model parameters:',get_n_params(model))
-    model.load_state_dict(torch.load(args.ckpt_dir))
-    model = model.to(device)
-
-    test_score, test_avg_precision,test_easy_avg_precision,test_hard_avg_precision = evaluate_segmentation(model, test_iterator, device,len(test_data),is_avg_prec=True,prec_thresholds=[0.5,0.6,0.7,0.8,0.9],output_dir=args.output_dir)
-    print('INFO: Dice score:', test_score)
-    print('INFO: Average precision at ordered thresholds:', test_avg_precision)
-    print('INFO: Easy samples average precision at ordered thresholds:', test_easy_avg_precision)
-    print('INFO: Hard samples average precision at ordered thresholds:', test_hard_avg_precision)
-
-if __name__ == "__main__":
-    ap = argparse.ArgumentParser()
-    ap.add_argument("--test_set_dir",required=True,type=str,help="path for the test dataset")
-    ap.add_argument("--ckpt_dir",required=True,type=str,help="path for the checkpoint of segmentation model to test")
-    ap.add_argument("--output_dir", required=True, type=str, help="path for saving the test outputs")
-
-    args = ap.parse_args()
-
-    assert os.path.isdir(args.test_set_dir), 'No such file or directory: ' + args.test_set_dir
-    if not os.path.isdir(os.path.join(args.output_dir,'input_segmentation_images')):
-        os.makedirs(os.path.join(args.output_dir,'input_segmentation_images'))
-    if not os.path.isdir(os.path.join(args.output_dir,'segmentation_predictions')):
-        os.makedirs(os.path.join(args.output_dir,'segmentation_predictions'))
-
-=======
 import torch.utils.data as data
 import segmentation_transforms as transforms
 import numpy as np
@@ -85,15 +22,15 @@ def test(args,image_size = [383,512],image_means = [0.5],image_stds= [0.5],batch
     device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
     test_transforms = transforms.Compose([
+                               transforms.ToPILImage(),
                                transforms.Resize(image_size),
-                               transforms.Grayscale(num_output_channels=1),
                                transforms.ToTensor(),
                                transforms.Normalize(mean = image_means,
                                                     std = image_stds)
                            ])
 
 
-    test_data = BasicSegmentationDataset(os.path.join(args.test_dir, 'images'), os.path.join(args.test_dir, 'masks'),os.path.join(args.test_dir, 'wmaps'),transforms=test_transforms)
+    test_data = BasicSegmentationDataset(os.path.join(args.test_set_dir, 'images'), os.path.join(args.test_set_dir, 'masks'),os.path.join(args.test_set_dir, 'wmaps'),transforms=test_transforms)
 
     test_iterator = data.DataLoader(test_data,batch_size = batch_size,shuffle=False)
 
@@ -110,17 +47,16 @@ def test(args,image_size = [383,512],image_means = [0.5],image_stds= [0.5],batch
 
 if __name__ == "__main__":
     ap = argparse.ArgumentParser()
-    ap.add_argument("--test_dir",required=True,type=str,help="path for the test dataset")
+    ap.add_argument("--test_set_dir",required=True,type=str,help="path for the test dataset")
     ap.add_argument("--ckpt_dir",required=True,type=str,help="path for the checkpoint of segmentation model to test")
     ap.add_argument("--output_dir", required=True, type=str, help="path for saving the test outputs")
 
     args = ap.parse_args()
 
-    assert os.path.isdir(args.test_dir), 'No such file or directory: ' + args.test_dir
-    if not os.path.isdir(args.output_dir+'/input_segmentation_images'):
-        os.makedirs(args.output_dir+'/input_segmentation_images')
-    if not os.path.isdir(args.output_dir+'/segmentation_predictions'):
-        os.makedirs(args.output_dir+'/segmentation_predictions')
+    assert os.path.isdir(args.test_set_dir), 'No such file or directory: ' + args.test_set_dir
+    if not os.path.isdir(os.path.join(args.output_dir,'input_segmentation_images')):
+        os.makedirs(os.path.join(args.output_dir,'input_segmentation_images'))
+    if not os.path.isdir(os.path.join(args.output_dir,'segmentation_predictions')):
+        os.makedirs(os.path.join(args.output_dir,'segmentation_predictions'))
 
->>>>>>> 688bc95bb88028284a86f389a4e204f93b8d0f83:deepsea/test_segmentation.py
-    test(args)
+    test(args)
diff --git a/deepsea/evaluate_test_set_tracking.py b/deepsea/evaluate_test_set_tracking.py
@@ -1,69 +1,3 @@
-<<<<<<< HEAD:evaluate_test_set_tracking.py
-import os
-import torch.utils.data as data
-import tracker_transforms as transforms
-import numpy as np
-import argparse
-import random
-from model import DeepSeaTracker
-from data import BasicTrackerDataset
-import torch
-from evaluate import evaluate_tracker
-from utils import get_n_params
-
-SEED = 1234
-random.seed(SEED)
-np.random.seed(SEED)
-torch.manual_seed(SEED)
-torch.cuda.manual_seed(SEED)
-torch.backends.cudnn.deterministic = True
-
-
-
-def test(args,image_size = [128,128],image_means = [0.5],image_stds= [0.5],batch_size=1):
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-
-    test_transforms = transforms.Compose([
-                               transforms.Resize(image_size),
-                               transforms.Grayscale(num_output_channels=1),
-                               transforms.ToTensor(),
-                               transforms.Normalize(mean = image_means,
-                                                    std = image_stds)
-                           ])
-
-
-    test_data = BasicTrackerDataset(os.path.join(args.test_set_dir), transforms=test_transforms,if_test=True)
-    test_iterator = data.DataLoader(test_data,batch_size = batch_size)
-
-    model=DeepSeaTracker(n_channels=1, n_classes=2, bilinear=True)
-    print('INFO: Num of model parameters:',get_n_params(model))
-
-    model.load_state_dict(torch.load(args.ckpt_dir))
-    model = model.to(device)
-
-    test_score, test_avg_precision,test_single_cell_avg_precision,test_mitosis_avg_precision = evaluate_tracker(model, test_iterator, device,len(test_data),is_avg_prec=True,prec_thresholds=[0.2,0.6,0.7,0.8,0.9],output_dir=args.output_dir)
-
-    print('INFO: Dice score:', test_score)
-    print('INFO: Average precision:', test_avg_precision)
-    print('INFO: Single cells average precision:', test_single_cell_avg_precision)
-    print('INFO: Mitosis average precision:', test_mitosis_avg_precision)
-
-
-if __name__ == "__main__":
-    ap = argparse.ArgumentParser()
-    ap.add_argument("--test_set_dir",required=True,type=str,help="path for the test dataset")
-    ap.add_argument("--ckpt_dir",required=True,type=str,help="path for the checkpoint of tracking model to test")
-    ap.add_argument("--output_dir", required=True, type=str, help="path for saving the test outputs")
-
-    args = ap.parse_args()
-
-    assert os.path.isdir(args.test_set_dir), 'No such file or directory: ' + args.test_set_dir
-    if not os.path.isdir(os.path.join(args.output_dir,'input_crops')):
-        os.makedirs(os.path.join(args.output_dir,'input_crops'))
-    if not os.path.isdir(os.path.join(args.output_dir,'tracking_predictions')):
-        os.makedirs(os.path.join(args.output_dir,'tracking_predictions'))
-
-=======
 import os
 import torch.utils.data as data
 import tracker_transforms as transforms
@@ -89,6 +23,7 @@ def test(args,image_size = [128,128],image_means = [0.5],image_stds= [0.5],batch
     device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
     test_transforms = transforms.Compose([
+                               transforms.ToPILImage(),
                                transforms.Resize(image_size),
                                transforms.Grayscale(num_output_channels=1),
                                transforms.ToTensor(),
@@ -97,7 +32,7 @@ def test(args,image_size = [128,128],image_means = [0.5],image_stds= [0.5],batch
                            ])
 
 
-    test_data = BasicTrackerDataset(os.path.join(args.test_dir), transforms=test_transforms,if_test=True)
+    test_data = BasicTrackerDataset(os.path.join(args.test_set_dir), transforms=test_transforms,if_test=True)
     test_iterator = data.DataLoader(test_data,batch_size = batch_size)
 
     model=DeepSeaTracker(n_channels=1, n_classes=2, bilinear=True)
@@ -116,17 +51,16 @@ def test(args,image_size = [128,128],image_means = [0.5],image_stds= [0.5],batch
 
 if __name__ == "__main__":
     ap = argparse.ArgumentParser()
-    ap.add_argument("--test_dir",required=True,type=str,help="path for the test dataset")
+    ap.add_argument("--test_set_dir",required=True,type=str,help="path for the test dataset")
     ap.add_argument("--ckpt_dir",required=True,type=str,help="path for the checkpoint of tracking model to test")
     ap.add_argument("--output_dir", required=True, type=str, help="path for saving the test outputs")
 
     args = ap.parse_args()
 
-    assert os.path.isdir(args.test_dir), 'No such file or directory: ' + args.test_dir
-    if not os.path.isdir(args.output_dir+'/input_tracking_images'):
-        os.makedirs(args.output_dir+'/input_tracking_images')
-    if not os.path.isdir(args.output_dir+'/tracking_predictions'):
-        os.makedirs(args.output_dir+'/tracking_predictions')
+    assert os.path.isdir(args.test_set_dir), 'No such file or directory: ' + args.test_set_dir
+    if not os.path.isdir(os.path.join(args.output_dir,'input_crops')):
+        os.makedirs(os.path.join(args.output_dir,'input_crops'))
+    if not os.path.isdir(os.path.join(args.output_dir,'tracking_predictions')):
+        os.makedirs(os.path.join(args.output_dir,'tracking_predictions'))
 
->>>>>>> 688bc95bb88028284a86f389a4e204f93b8d0f83:deepsea/test_tracker.py
     test(args)
diff --git a/deepsea/measure_MOTA.py b/deepsea/measure_MOTA.py
@@ -51,8 +51,8 @@ def main(args,seg_img_size= [383,512],tracking_image_size = [128,128],image_mean
 
     device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
     seg_transforms = segmentation_transforms.Compose([
+                               segmentation_transforms.ToPILImage(),
                                segmentation_transforms.Resize(seg_img_size),
-                               segmentation_transforms.Grayscale(num_output_channels=1),
                                segmentation_transforms.ToTensor(),
                                segmentation_transforms.Normalize(mean = image_means,
                                                     std = image_stds)
@@ -99,7 +99,6 @@ def main(args,seg_img_size= [383,512],tracking_image_size = [128,128],image_mean
     ap.add_argument("--output_dir", required=True, type=str, help="path for saving the test outputs")
 
     args = ap.parse_args()
-
     assert os.path.isdir(args.single_test_set_dir), 'No such file or directory: ' + args.single_test_set_dir
 
     if os.path.isdir(args.output_dir):