Updated GUI to allow for selecting custom models. Minor Bug Fixes

hcat/backends/detection.py

@@ -30,31 +30,31 @@ def init_model():
     return HairCellConvNext
 
 
-def FasterRCNN_from_url(url: str, device: str):
+def FasterRCNN_from_url(url: str,  device: str, path: str = None):
     """
     Loads a FasterRCNN model from a url OR from a local source if available.
 
     :param url: URL of pretrained model path. Will save the model to the source directory of HCAT.
     :param device: Device to load the model to.
     :return:
     """
-    path = os.path.join(hcat.__path__[0], 'detection_trained_model.trch')
-    print(path)
 
-    # Research Purposes...
-    # convnext = '/home/chris/Dropbox (Partners HealthCare)/HairCellInstance/Max_project_detection_resnet_NeXT101_04March_2022_MinValidation.trch'
-    # convnext = '/home/chris/Dropbox (Partners HealthCare)/trainHairCellDetection/models/Apr05_09-52-47_CHRISUBUNTU.trch'
-    # convnext = '/home/chris/Dropbox (Partners HealthCare)/trainHairCellDetection/models/Apr25_10-08-35_CHRISUBUNTU.trch'
-    # convnext = '/home/chris/Dropbox (Partners HealthCare)/trainHairCellDetection/models/Jul15_17-25-42_CHRISUBUNTU.trch'
-    # convnext_nocommunitydata = '/media/DataStorage/Dropbox (Partners HealthCare)/HairCellInstance/models/Max_project_detection_resnet_NeXT101_28January_2022_eFinal_just_arturlabtrainingdata.trch'
+    if path is None:  # we dont have a user defined file...
+        path = os.path.join(hcat.__path__[0], 'detection_trained_model.trch')
+        if not os.path.exists(path):
+            wget.download(url=url, out=path) # this will download the file...
 
+    # Check if the path exists from the user...
     if not os.path.exists(path):
-        wget.download(url=url, out=path)
+        raise RuntimeError(f'Could not locate the file at path: {path}')
 
     model = init_model()
 
-    print(path)
-    checkpoint = torch.load(path, map_location=torch.device('cpu'))
+    try:
+        checkpoint = torch.load(path, map_location=torch.device('cpu'))
+    except:
+        raise RuntimeError(f'Could not load torch model from file: {path}')
+
     if 'model_state_dict' in checkpoint:
         checkpoint = checkpoint['model_state_dict']
     model = model.to(device)

hcat/detect.py

@@ -189,7 +189,7 @@ def _detect(f: Optional[str] = None,
                  'limited. Consider Manual Calculation.', color='yellow')
 
         # curvature estimation really only works if there is a lot of tissue...
-        if distance is not None and distance.max() > 4000:
+        if distance is not None and distance.max() > 100:
             for c in cells: c.calculate_frequency(curvature[[0, 1], :], distance)  # calculate cell's best frequency
             cells = [c for c in cells if not c._distance_is_far_away]  # remove a cell if its far away from curve
 

hcat/detect_gui.py

@@ -33,13 +33,14 @@ class gui:
     def __init__(self):
         sg.theme('DarkGrey5')
         plt.ioff()
+        device_str = 'CUDA' if torch.cuda.is_available() else 'CPU'
         # sg.set_options(font='Any')
 
         button_column = [
             [sg.FileBrowse(size=(16, 1), enable_events=True), ],
             [sg.Button('Load', size=(16, 1)), ],
             [sg.Button('Save', size=(16, 1))],
-            [sg.Button("⇦",k="previous_image", size=(4, 1)), sg.Push(), sg.Button("⇨", k="next_image", size=(4,1))],
+            [sg.Button("⇦", k="previous_image", size=(4, 1)), sg.Push(), sg.Button("⇨", k="next_image", size=(4,1))],
             [sg.HorizontalSeparator(p=(0, 20))],
             [sg.Text('Cell Diameter\n(In Pixels)')],
             [sg.Input(size=(10, 1), enable_events=True, default_text=30, key='Diameter'),
@@ -49,6 +50,12 @@ def __init__(self):
             [sg.Text('Overlap Threshold')],
             [sg.Slider(range=(0, 100), orientation='h', enable_events=True, default_value=30, key='NMS', expand_x=True)],
             [sg.HorizontalSeparator(p=(0, 20))],
+            [sg.Text('Model Selection', size=(15, 1), pad=(0, 10))],
+            [sg.FileBrowse(button_text='Select', k='select_model',
+                           size=(15, 1), enable_events=True, file_types=('Pytorch Model', '*.trch'))],
+            [sg.Button('Reset', k='reset_model', size=(15,1))],
+            [sg.Text('Default', k='model_selection_text', size=(15, 1))],
+            [sg.HorizontalSeparator(p=(0, 20))],
             [sg.Button('Run Analysis', size=(15, 1))],
             [sg.Check(text=' Live Update', key='live_update', enable_events=True)],
             [sg.Check(k='savexml', text=' Save XML', default=False, enable_events=True)]
@@ -96,6 +103,7 @@ def __init__(self):
             [sg.HorizontalSeparator(pad=(0, 30))],
             [sg.Text('OHC: None', key='OHC_count')],
             [sg.Text('IHC: None', key='IHC_count')],
+            [sg.Text(f'Hardware Accelerator: {device_str}', key='device')]
         ]
 
         layout = [[sg.Column(button_column, vertical_alignment='Top'),
@@ -141,7 +149,10 @@ def __init__(self):
         self.rgb = None
         self.contrast = None
         self.brightness = None
+
         self.model = None
+        self.model_path = None
+
         self.device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
         self.cochlea = None
 
@@ -153,7 +164,6 @@ def __init__(self):
         self.labels = None
         self.scores = None
 
-        self.model = None
 
 
     def main_loop(self):
@@ -171,6 +181,22 @@ def main_loop(self):
             if event == 'pan':
                 print('CLICKED: ', values['pan'])
 
+            if event == 'select_model':
+                self.model_path = values['select_model']
+                self.window['model_selection_text'].update(f'Custom')
+                try:
+                    self.model = FasterRCNN_from_url(url=None, device=self.device, path=self.model_path)
+                except:
+                    sg.popup_ok('Could not load the model from file.')
+                    self.model_path = None
+                    self.window['model_selection_text'].update(f'Default')
+
+            if event == 'reset_model':
+                self.model_path = None
+                self.model = None
+                self.window['model_selection_text'].update(f'Default')
+
+
             if event == 'Exit' or event == sg.WIN_CLOSED:
                 return
 
@@ -202,7 +228,6 @@ def main_loop(self):
                         self.current_image_index = i
                         break
 
-
                 if self.valid_image_files[self.current_image_index] != values['Browse']:
                     print(f'{self.current_image_index}, {self.valid_image_files}')
 
@@ -283,8 +308,8 @@ def main_loop(self):
 
             if event == 'Run Analysis' and self.__LOADED__:
                 # self.run_detection_model()
-                self.fast_model()
-                self.threshold_and_nms(values['Threshold'], values['NMS'])
+                self.fast_model() # runs model on image...
+                self.threshold_and_nms(values['Threshold'], values['NMS']) # removing junk data
                 self.draw_image()
 
             if event in ['Threshold', 'NMS']:
@@ -566,7 +591,11 @@ def fast_model(self):
         if self.model is None:
             __model_url__ = 'https://www.dropbox.com/s/opf43jwcbgz02vm/detection_trained_model.trch?dl=1'
             sg.popup_quick_message('Loading model from file. May take a while.')
-            self.model = FasterRCNN_from_url(url=__model_url__, device=self.device)
+            try:
+                self.model = FasterRCNN_from_url(url=__model_url__, device=self.device)
+            except:
+                sg.popup_ok('Could not load the model from file.')
+                return None
 
         _image = self.scaled_image.clone()[0:3,...]
 

hcat/lib/cochlea.py

@@ -249,7 +249,7 @@ def write_csv(self, filename: Optional[Union[bool, str]] = None) -> None:
         :return: None
         """
         if self.analysis_type == 'segment':
-            label = 'cellID,frequency,percent_loc,x_loc,y_loc,z_loc,volume,summed,'
+            label = 'cellID,distance,frequency,percent_loc,x_loc,y_loc,z_loc,volume,summed,'
             for c in ['myo', 'dapi', 'actin', 'gfp']:
                 label += f'{c}_mean,{c}_median,{c}_std,{c}_var,{c}_min,{c}_max,{c}_%zero,{c}_%saturated,'
 
@@ -266,7 +266,7 @@ def write_csv(self, filename: Optional[Union[bool, str]] = None) -> None:
             f.write(label[:-1:] + '\n')  # index to remove final comma
 
             for cell in self.cells:
-                f.write(f'{cell.id},{cell.frequency},{cell.percent_loc},')
+                f.write(f'{cell.id},{cell.distance},{cell.frequency},{cell.percent_loc},')
                 f.write(f'{cell.loc[1]},{cell.loc[2]},{cell.loc[3]},{cell.volume},{cell.summed},')
 
                 for id in cell.channel_names:
@@ -277,7 +277,7 @@ def write_csv(self, filename: Optional[Union[bool, str]] = None) -> None:
                 f.write('\n')
             f.close()
         elif self.analysis_type == 'detect':
-            label = 'cellID,type,score,frequency,percent_loc,x_loc,y_loc'
+            label = 'cellID,type,score,distance,frequency,percent_loc,x_loc,y_loc'
 
             if filename is None and self.path is not None:
                 filename = os.path.splitext(self.path)[0] + '.csv'  # Remove .lif and add .csv
@@ -292,7 +292,7 @@ def write_csv(self, filename: Optional[Union[bool, str]] = None) -> None:
             f.write(label[:-1:] + '\n')  # index to remove final comma
 
             for cell in self.cells:
-                f.write(f'{cell.id},{cell.type},{cell.scores},{cell.frequency},{cell.percent_loc},')
+                f.write(f'{cell.id},{cell.type},{cell.scores},{cell.distance},{cell.frequency},{cell.percent_loc},')
                 f.write(f'{cell.loc[1]},{cell.loc[2]}')
                 f.write('\n')
             f.close()

hcat/lib/functional.py

@@ -146,7 +146,12 @@ def __call__(self,
             3) The estimated apex of the cochlea
         """
         curvature, distance, apex = None, None, None
-        if self.method is None:
+
+        if csv is not None:
+            print('RUNNING EPL METHOD')
+            curvature, distance, apex = self.fitEPL(csv)
+
+        elif self.method is None:
             if csv is not None:
                 curvature, distance, apex = self.fitEPL(csv)
 

hcat/lib/utils.py

@@ -295,6 +295,8 @@ def load(file: str, header_name: Optional[str] = 'TileScan 1 Merged',
             pass
         elif image_base.ndim == 3 and image_base.shape[0] <= 4:  # Suppose you load a 2D image! with multiple channels
             pass
+        elif image_base.ndim == 2:
+            image_base = image_base[np.newaxis, ...]
         else:
             print(
                 f'\x1b[1;31;40m' + f'Cannot load: \'{file}\'. Unsupported number of dimmensions: {image_base.ndim}' + '\x1b[0m')
@@ -478,7 +480,7 @@ def cochlea_to_xml(cochlea, filename: str) -> None:
     tree.write(filename + '.xml')
 
 
-def normalize_image(image: Tensor, verbose: Optional[bool] = False) -> Tensor:
+def normalize_image(image: Tensor, *args, verbose: Optional[bool] = False) -> Tensor:
     """
     Normalizes each channel in an image such that a majority of the image lies between 0 and 1.
     Calculates the maximum value of the image following a gaussian blur with a 7x7 kernel size,

hcat/main.py

@@ -1,5 +1,6 @@
 from hcat.detect import _detect
 from hcat.detect_gui import gui
+import os.path
 
 import glob
 
@@ -32,7 +33,7 @@ def cli(ctx):
 @click.option('--normalize', is_flag=True, help='Threshold (between 0 and 1) of cell detection.')
 @click.option('--pixel_size', default=None, help='Pixel size in nm')
 @click.option('--cell_diameter', default=None, help='Cell diameter in pixels')
-@click.option('--predict_curvature', default=None, help='Cell diameter in pixels')
+@click.option('--predict_curvature', is_flag=True, help='Cell diameter in pixels')
 @click.option('--silent', default=False, help="Suppresses most of HCAT's logging ")
 def detect(f: str, curve_path, cell_detection_threshold, nms_threshold, save_xml, save_png,
            save_fig, normalize, pixel_size, dtype, cell_diameter, predict_curvature, silent):
@@ -44,6 +45,12 @@ def detect(f: str, curve_path, cell_detection_threshold, nms_threshold, save_xml
 
     files = glob.glob(f)
     for filename in files:
+        curve_path = filename[:-4:] + '_path.csv'
+        print('CURVE PATH: ', curve_path)
+
+        if not os.path.exists(filename[:-4:] + '_path.csv'):
+            raise ValueError(curve_path)
+
         try:
             _detect(f=filename,
                     curve_path=curve_path,

setup.cfg

@@ -1,6 +1,6 @@
 [metadata]
 name = hcat
-version = 0.2.24
+version = 0.2.26
 author = Chris Buswinka
 author_email = [email protected]
 classifiers = Programming Language :: Python :: 3