megadetector 5.0.5__py3-none-any.whl → 5.0.7__py3-none-any.whl

md_utils/write_html_image_list.py

@@ -8,16 +8,13 @@
 # Each "filename" can also be a dict with elements 'filename','title',
 # 'imageStyle','textStyle', 'linkTarget'
 #
-# Strips directory information away if options.makeRelative == 1.
-#
-# Tries to convert absolute to relative paths if options.makeRelative == 2.
-#
 ########
 
 #%% Constants and imports
 
 import os
 import math
+import urllib
 
 from md_utils import path_utils
 
@@ -39,12 +36,12 @@ def write_html_image_list(filename=None,images=None,options=None):
     options: a dict with one or more of the following fields:
         
         fHtml
-        makeRelative
         headerHtml
         trailerHtml
         defaultTextStyle
         defaultImageStyle
         maxFiguresPerHtmlFile
+        urlEncodeFilenames (default True, e.g. '#' will be replaced by '%23')
         
     """
     
@@ -55,9 +52,6 @@ def write_html_image_list(filename=None,images=None,options=None):
     if 'fHtml' not in options:
         options['fHtml'] = -1
     
-    if 'makeRelative' not in options:
-        options['makeRelative'] = 0
-    
     if 'headerHtml' not in options or options['headerHtml'] is None:
         options['headerHtml'] = ''        
     
@@ -71,6 +65,9 @@ def write_html_image_list(filename=None,images=None,options=None):
     if 'defaultImageStyle' not in options or options['defaultImageStyle'] is None:
         options['defaultImageStyle'] = \
         "margin:0px;margin-top:5px;margin-bottom:5px;"
+    
+    if 'urlEncodeFilenames' not in options or options['urlEncodeFilenames'] is None:
+        options['urlEncodeFilenames'] = True
         
     # Possibly split the html output for figures into multiple files; Chrome gets sad with
     # thousands of images in a single tab.        
@@ -98,24 +95,6 @@ def write_html_image_list(filename=None,images=None,options=None):
             imageInfo['textStyle'] = options['defaultTextStyle']
         images[iImage] = imageInfo            
     
-    # Remove leading directory information from filenames if requested
-    if options['makeRelative'] == 1:
-        
-        for iImage in range(0,len(images)):
-            _,n,e = path_utils.fileparts(images[iImage]['filename'])
-            images[iImage]['filename'] = n + e
-        
-    elif options['makeRelative'] == 2:
-        
-        baseDir,_,_ = path_utils.fileparts(filename)
-        if len(baseDir) > 1 and baseDir[-1] != '\\':
-            baseDir = baseDir + '\\'
-        
-        for iImage in range(0,len(images)):
-            fn = images[iImage]['filename']
-            fn = fn.replace(baseDir,'')
-            images[iImage]['filename'] = fn        
-    
     nImages = len(images)
     
     # If we need to break this up into multiple files...
@@ -197,6 +176,10 @@ def write_html_image_list(filename=None,images=None,options=None):
         title = title.encode('ascii','ignore').decode('ascii')
         filename = filename.encode('ascii','ignore').decode('ascii')
         
+        if options['urlEncodeFilenames']:
+            filename = filename.replace('\\','/')
+            filename = urllib.parse.quote(filename)
+        
         if len(title) > 0:       
             fHtml.write(
                     '<p style="{}">{}</p>\n'\

md_visualization/plot_utils.py

@@ -28,8 +28,9 @@ def plot_confusion_matrix(
         cmap: Union[str, matplotlib.colors.Colormap] = matplotlib.cm.Blues,
         vmax: Optional[float] = None,
         use_colorbar: bool = True,
-        y_label: bool = True,
-        fmt: str = '{:.0f}'
+        y_label: bool = True,        
+        fmt: str = '{:.0f}',
+        fig=None
         ) -> matplotlib.figure.Figure:
     """
     Plot a confusion matrix. By default, assumes values in the given matrix
@@ -56,18 +57,20 @@ def plot_confusion_matrix(
     assert matrix.shape[1] == num_classes
     assert len(classes) == num_classes
 
+    normalized_matrix = matrix.astype(np.float64) / (
+        matrix.sum(axis=1, keepdims=True) + 1e-7)
     if normalize:
-        matrix = matrix.astype(np.float64) / (
-            matrix.sum(axis=1, keepdims=True) + 1e-7)
+        matrix = normalized_matrix
 
     fig_h = 3 + 0.3 * num_classes
     fig_w = fig_h
     if use_colorbar:
         fig_w += 0.5
 
-    fig = matplotlib.figure.Figure(figsize=(fig_w, fig_h), tight_layout=True)
+    if fig is None:
+        fig = matplotlib.figure.Figure(figsize=(fig_w, fig_h), tight_layout=True)
     ax = fig.subplots(1, 1)
-    im = ax.imshow(matrix, interpolation='nearest', cmap=cmap, vmax=vmax)
+    im = ax.imshow(normalized_matrix, interpolation='nearest', cmap=cmap, vmax=vmax)
     ax.set_title(title)
 
     if use_colorbar:
@@ -86,10 +89,11 @@ def plot_confusion_matrix(
         ax.set_ylabel('Ground-truth class')
 
     for i, j in np.ndindex(matrix.shape):
-        ax.text(j, i, fmt.format(matrix[i, j] * 100),
+        v = matrix[i, j]        
+        ax.text(j, i, fmt.format(v),
                 horizontalalignment='center',
                 verticalalignment='center',
-                color='white' if matrix[i, j] > 0.5 else 'black')
+                color='white' if normalized_matrix[i, j] > 0.5 else 'black')
 
     return fig
 

md_visualization/visualization_utils.py

@@ -166,13 +166,26 @@ def load_image(input_file: Union[str, BytesIO]) -> Image:
     return image
 
 
-def resize_image(image, target_width, target_height=-1):
+def resize_image(image, target_width, target_height=-1, output_file=None):
     """
     Resizes a PIL image object to the specified width and height; does not resize
     in place. If either width or height are -1, resizes with aspect ratio preservation.
     If both are -1, returns the original image (does not copy in this case).
+    
+    None is equivalent to -1 for target_width and target_height.
+    
+    [image] can be a PIL image or a filename.
     """
 
+    if isinstance(image,str):
+        image = load_image(image)
+        
+    if target_width is None:
+        target_width = -1
+    
+    if target_height is None:
+        target_height = -1
+        
     # Null operation
     if target_width == -1 and target_height == -1:
         return image
@@ -197,6 +210,9 @@ def resize_image(image, target_width, target_height=-1):
     except:
         resized_image = image.resize((target_width, target_height), Image.Resampling.LANCZOS)
         
+    if output_file is not None:
+        exif_preserving_save(resized_image,output_file)
+        
     return resized_image
 
 
@@ -363,7 +379,8 @@ def render_detection_bounding_boxes(detections, image,
             The type of the numerical label (default string) needs to be consistent with the keys in
             label_map; no casting is carried out.  If this is None, no classification labels are shown.
 
-        confidence_threshold: optional, threshold above which the bounding box is rendered.
+        confidence_threshold: optional, threshold above which boxes are rendered.  Can also be a dictionary
+        mapping category IDs to thresholds.
         
         thickness: line thickness in pixels. Default value is 4.
         
@@ -397,9 +414,15 @@ def render_detection_bounding_boxes(detections, image,
 
         score = detection['conf']
         
+        if isinstance(confidence_threshold,dict):
+            rendering_threshold = confidence_threshold[detection['category']]
+        else:
+            rendering_threshold = confidence_threshold        
+            
+            
         # Always render objects with a confidence of "None", this is typically used
         # for ground truth data.        
-        if score is None or score >= confidence_threshold:
+        if score is None or score >= rendering_threshold:
             
             x1, y1, w_box, h_box = detection['bbox']
             display_boxes.append([y1, x1, y1 + h_box, x1 + w_box])
@@ -784,7 +807,8 @@ def draw_bounding_boxes_on_file(input_file, output_file, detections, confidence_
                                 detector_label_map=DEFAULT_DETECTOR_LABEL_MAP,
                                 thickness=DEFAULT_BOX_THICKNESS, expansion=0,
                                 colormap=DEFAULT_COLORS,
-                                custom_strings=None):
+                                label_font_size=DEFAULT_LABEL_FONT_SIZE,
+                                custom_strings=None,target_size=None):
     """
     Render detection bounding boxes on an image loaded from file, writing the results to a
     new image file.
@@ -804,15 +828,21 @@ def draw_bounding_boxes_on_file(input_file, output_file, detections, confidence_
     custom_strings: optional set of strings to append to detection labels, should have the
     same length as [detections].  Appended before classification labels, if classification
     data is provided.
+    
+    target_size: tuple of (target_width,target_height).  Either or both can be -1,
+    see resize_image for documentation.  If None or (-1,-1), uses the original image size.
     """
     
     image = open_image(input_file)
-
+    
+    if target_size is not None:
+        image = resize_image(image,target_size[0],target_size[1])
+        
     render_detection_bounding_boxes(
             detections, image, label_map=detector_label_map,
             confidence_threshold=confidence_threshold,
             thickness=thickness,expansion=expansion,colormap=colormap,
-            custom_strings=custom_strings)
+            custom_strings=custom_strings,label_font_size=label_font_size)
 
     image.save(output_file)
 
@@ -838,4 +868,73 @@ def draw_db_boxes_on_file(input_file, output_file, boxes, classes=None,
 
     image.save(output_file)
     
-    
+
+def gray_scale_fraction(image,crop_size=(0.1,0.1)):
+    """
+    Returns the fraction of the pixels in [image] that appear to be grayscale (R==G==B), 
+    useful for approximating whether this is a night-time image when flash information is not
+    available in EXIF data (or for video frames, where this information is often not available
+    in structured metadata at all).
+    
+    [image] can be a PIL image or a file name.
+    
+    crop_size should be a 2-element list/tuple, representing the fraction of the image 
+    to crop at the top and bottom, respectively, before analyzing (to minimize the possibility
+    of including color elements in the image chrome).
+    """
+    
+    if isinstance(image,str):
+        image = Image.open(image)
+    
+    if image.mode == 'L':
+        return 1.0
+    
+    if len(image.getbands()) == 1:
+        return 1.0
+    
+    # Crop if necessary
+    if crop_size[0] > 0 or crop_size[1] > 0:
+        
+        assert (crop_size[0] + crop_size[1]) < 1.0, \
+            print('Illegal crop size: {}'.format(str(crop_size)))
+            
+        top_crop_pixels = int(image.height * crop_size[0])
+        bottom_crop_pixels = int(image.height * crop_size[1])
+        
+        left = 0
+        right = image.width
+        
+        # Remove pixels from the top
+        first_crop_top = top_crop_pixels
+        first_crop_bottom = image.height        
+        first_crop = image.crop((left, first_crop_top, right, first_crop_bottom))
+        
+        # Remove pixels from the bottom
+        second_crop_top = 0
+        second_crop_bottom = first_crop.height - bottom_crop_pixels
+        second_crop = first_crop.crop((left, second_crop_top, right, second_crop_bottom))
+        
+        image = second_crop
+    
+    # It doesn't matter if these are actually R/G/B, they're just names
+    r = np.array(image.getchannel(0))
+    g = np.array(image.getchannel(1))
+    b = np.array(image.getchannel(2))
+        
+    gray_pixels = np.logical_and(r == g, r == b)
+    n_pixels = gray_pixels.size
+    n_gray_pixels = gray_pixels.sum()
+    
+    return n_gray_pixels / n_pixels
+
+    # Non-numpy way to do the same thing, briefly keeping this here for posterity
+    if False:
+        
+        w, h = image.size
+        n_pixels = w*h
+        n_gray_pixels = 0
+        for i in range(w):
+            for j in range(h):
+                r, g, b = image.getpixel((i,j))
+                if r == g and r == b and g == b:
+                    n_gray_pixels += 1            

md_visualization/visualize_db.py

@@ -3,7 +3,7 @@
 # visualize_db.py
 # 
 # Outputs an HTML page visualizing annotations (class labels and/or bounding boxes)
-# on a sample of images in a database in the COCO Camera Traps format
+# on a sample of images in a database in the COCO Camera Traps format.
 #
 ########
 
@@ -41,7 +41,15 @@ class DbVizOptions:
     #
     # If viz_size is None or (-1,-1), the original image size is used.
     viz_size = (675, -1)
+    
+    # The most relevant option one might want to set here is:
+    #
+    # htmlOptions['maxFiguresPerHtmlFile']
+    #
+    # ...which can be used to paginate previews to a number of images that will load well
+    # in a browser (5000 is a reasonable limit).
     htmlOptions = write_html_image_list()
+    
     sort_by_filename = True
     trim_to_images_with_bboxes = False
     
@@ -96,7 +104,7 @@ def image_filename_to_path(image_file_name, image_base_dir, pathsep_replacement=
 
 #%% Core functions
 
-def process_images(db_path, output_dir, image_base_dir, options=None):
+def visualize_db(db_path, output_dir, image_base_dir, options=None):
     """
     Writes images and html to output_dir to visualize the annotations in the json file
     db_path.
@@ -418,7 +426,7 @@ def process_images(db_path, output_dir, image_base_dir, options=None):
     
     return htmlOutputFile,image_db
 
-# def process_images(...)
+# def visualize_db(...)
     
     
 #%% Command-line driver
@@ -452,7 +460,8 @@ def main():
     parser.add_argument('--random_seed', action='store', type=int, default=None,
                         help='Random seed for image selection')
     parser.add_argument('--pathsep_replacement', action='store', type=str, default='',
-                        help='Replace path separators in relative filenames with another character (frequently ~)')
+                        help='Replace path separators in relative filenames with another ' + \
+                             'character (frequently ~)')
     
     if len(sys.argv[1:]) == 0:
         parser.print_help()
@@ -466,7 +475,7 @@ def main():
     if options.random_sort:
         options.sort_by_filename = False
         
-    process_images(options.db_path,options.output_dir,options.image_base_dir,options) 
+    visualize_db(options.db_path,options.output_dir,options.image_base_dir,options) 
 
 
 if __name__ == '__main__':
@@ -474,7 +483,7 @@ if __name__ == '__main__':
     main()
 
 
-#%% Interactive driver(s)
+#%% Interactive driver
 
 if False:
     
@@ -487,6 +496,5 @@ if False:
     options = DbVizOptions()
     options.num_to_visualize = 100
     
-    htmlOutputFile,db = process_images(db_path,output_dir,image_base_dir,options)
+    htmlOutputFile,db = visualize_db(db_path,output_dir,image_base_dir,options)
     # os.startfile(htmlOutputFile)
-