from PIL import ImageDraw, ImageFont, Image import numpy as np import cv2 def _get_text_dimensions(draw, text, font): """ Calculate the width and height of text in pixels. Args: draw: ImageDraw object text: String of text to measure font: ImageFont object Returns: tuple: (width, height) in pixels """ # Get the bounding box of the text bbox = draw.textbbox((0, 0), text, font=font) # Calculate width and height from the bounding box # bbox returns (left, top, right, bottom) width = bbox[2] - bbox[0] height = bbox[3] - bbox[1] return width, height def add_landmarks_and_line_overlay(pil_img, p1_coords, p2_coords, line_color="#00FF00", point_color="#FF0000"): """ Add landmarks (points) and a line connecting them to an image. Args: pil_img: PIL Image in RGB or RGBA format p1_coords: List of [dim0, dim1] coordinates for the first point p2_coords: List of [dim0, dim1] coordinates for the second point Returns: PIL Image with the landmarks and connecting line overlay """ # Create a drawing object draw = ImageDraw.Draw(pil_img) # Convert coordinates format (similar to bbox function) # First coordinate is height (y) and second is width (x) x1, y1 = p1_coords[1], p1_coords[0] x2, y2 = p2_coords[1], p2_coords[0] # Draw the line connecting the points (green) draw.line([(x1, y1), (x2, y2)], fill=line_color, width=2) # Draw the points (red) point_radius = 3 draw.ellipse( [ (x1 - point_radius, y1 - point_radius), (x1 + point_radius, y1 + point_radius), ], fill=point_color, ) draw.ellipse( [ (x2 - point_radius, y2 - point_radius), (x2 + point_radius, y2 + point_radius), ], fill=point_color, ) return pil_img def add_bbox_overlay(pil_img, bbox_min_coords, bbox_max_coords): """ Add a bounding box overlay to an image. Args: pil_img: PIL Image in RGB or RGBA format bbox_min_coords: List of [dim0_min, dim1_min] coordinates for the top-left corner of the bounding box bbox_max_coords: List of [dim0_max, dim1_max] coordinates for the bottom-right corner of the bounding box NOTE: For the coordinate definition in the MedVision dataset, please refer to the `medvision_ds.utils.benchmark_planner.MedVision_BenchmarkPlannerDetection._find_bounding_boxes_2D` Returns: PIL Image with the bounding box overlay """ # NOTE: For bbox_min_coords and bbox_max_coords: # the first coordinate is the height (y-axis) direction and the second is the width (x-axis) direction; # the origin is at the upper-left corner of the image. # For PIL Image, the origin is at the upper-left corner of the image. # So, x-coordinate = dim1_coordinate, y-coordinate = dim0_coordinate # Convert input bounding box corrdinates to xy coordinates for PIL Image x_min = bbox_min_coords[1] y_min = bbox_min_coords[0] x_max = bbox_max_coords[1] y_max = bbox_max_coords[0] # ref: https://pillow.readthedocs.io/en/stable/reference/ImageDraw.html draw = ImageDraw.Draw(pil_img) draw.rectangle([(x_min, y_min), (x_max, y_max)], outline="#00FF00", width=2) return pil_img def add_bbox_overlay_solid(pil_img, bbox_min_coords, bbox_max_coords): """ Add a semi-transparent solid bounding box overlay to an image. Args: pil_img: PIL Image in RGB or RGBA format bbox_min_coords: List of [dim0_min, dim1_min] coordinates for the top-left corner of the bounding box bbox_max_coords: List of [dim0_max, dim1_max] coordinates for the bottom-right corner of the bounding box NOTE: For the coordinate definition in the MedVision dataset, please refer to the `medvision_ds.utils.benchmark_planner.MedVision_BenchmarkPlannerDetection._find_bounding_boxes_2D` Returns: PIL Image with the solid bounding box overlay """ # NOTE: For bbox_min_coords and bbox_max_coords: # the first coordinate is the height (y-axis) direction and the second is the width (x-axis) direction; # the origin is at the upper-left corner of the image. # For PIL Image, the origin is at the upper-left corner of the image. # So, x-coordinate = dim1_coordinate, y-coordinate = dim0_coordinate # Convert input bounding box corrdinates to xy coordinates for PIL Image x_min = bbox_min_coords[1] y_min = bbox_min_coords[0] x_max = bbox_max_coords[1] y_max = bbox_max_coords[0] # Convert to RGBA mode to support transparency pil_img = pil_img.convert("RGBA") # ref: https://pillow.readthedocs.io/en/stable/reference/ImageDraw.html draw = ImageDraw.Draw(pil_img) # Draw a filled rectangle with semi-transparent green (25% opacity) draw.rectangle([(x_min, y_min), (x_max, y_max)], fill=(0, 255, 0, 64)) return pil_img def add_mask_overlay_contour(pil_img, mask_2d_binary): """ Add a green contour outline to an image based on a binary mask. Args: pil_img: PIL Image in RGB or RGBA format mask_2d_binary: Binary numpy array representing the mask Returns: PIL Image with the mask contour overlaid in green """ # Convert PIL image to numpy array for OpenCV img_np = np.array(pil_img) # Make sure mask is the right type and size mask = mask_2d_binary.astype(np.uint8) if mask.shape[:2] != img_np.shape[:2]: mask = cv2.resize( mask, (img_np.shape[1], img_np.shape[0]), interpolation=cv2.INTER_NEAREST ) # Find contours in the mask contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # Create a copy of the image to draw on img_with_contour = img_np.copy() # Draw contours on the image cv2.drawContours(img_with_contour, contours, -1, (0, 255, 0), 2) # Convert back to PIL return Image.fromarray(img_with_contour) def add_mask_overlay_solid(pil_img, mask_2d_binary): """ Add a semi-transparent green overlay to an image based on a binary mask. Args: pil_img: PIL Image in RGB or RGBA format mask_2d_binary: Binary numpy array representing the mask Returns: PIL Image with the mask overlaid in green """ # Create a green overlay image overlay = Image.new("RGBA", pil_img.size, (0, 255, 0, 0)) # Convert mask to PIL image format and resize if needed mask_pil = Image.fromarray((mask_2d_binary * 64).astype(np.uint8), mode="L") if mask_pil.size != pil_img.size: mask_pil = mask_pil.resize(pil_img.size) # Set the mask as the alpha channel for the overlay overlay.putalpha(mask_pil) # Convert original image to RGBA pil_img = pil_img.convert("RGBA") # Composite the images pil_img = Image.alpha_composite(pil_img, overlay) # Convert back to RGB for display pil_img = pil_img.convert("RGB") return pil_img def add_scale_label(pil_img, pixel_sizes, slice_dim): """Add scale label to image.""" draw = ImageDraw.Draw(pil_img) # Get image dimensions - in PIL, size returns (width, height) img_width, img_height = pil_img.size # Define a class with the _get_appropriate_scale method class ScaleCalculator: def _get_appropriate_scale(self, pixel_size, img_size, init_scale=10): scales = [1, 2, 5, 10, 15, 20, 25, 30, 40, 50, 60, 70, 80, 90, 100] scale_pixels_num = int(init_scale / pixel_size) min_pixels = img_size * 0.05 max_pixels = img_size * 0.25 if scale_pixels_num < min_pixels: for scale in scales: if scale > init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) elif scale_pixels_num > max_pixels: for scale in reversed(scales): if scale < init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) return init_scale, scale_pixels_num scale_calculator = ScaleCalculator() # Find which dimension is smaller # In the 2D array: height = first dimension, width = second dimension # In pixel_sizes: [height_scale, width_scale] # In PIL image: img_width = second dimension, img_height = first dimension if img_height < img_width: # Height is the smaller dimension pixel_size_min = pixel_sizes[0] # Height pixel size image_dim_min = img_height else: # Width is the smaller dimension pixel_size_min = pixel_sizes[1] # Width pixel size image_dim_min = img_width # Calculate appropriate scale scale_mm, scale_pixels_min = scale_calculator._get_appropriate_scale( pixel_size_min, image_dim_min, init_scale=10 ) # Calculate scale for the other dimension if img_height < img_width: scale_pixels_height = scale_pixels_min scale_pixels_width = int(scale_mm / pixel_sizes[1]) else: scale_pixels_width = scale_pixels_min scale_pixels_height = int(scale_mm / pixel_sizes[0]) # Position for scale bar (5% from the edge) start_x, start_y = int(img_width * 0.05), int(img_height * 0.05) end_x, end_y = start_x + scale_pixels_width, start_y + scale_pixels_height # Set text font and scale line width default_line_width = 2 default_fontsize = 10 line_width = default_line_width font = ImageFont.load_default().font_variant(size=default_fontsize) # Draw horizontal line draw.line( [(start_x, start_y), (end_x, start_y)], fill=(255, 255, 255), width=line_width, ) # Draw vertical line draw.line( [(start_x, start_y), (start_x, end_y)], fill=(255, 255, 255), width=line_width, ) # Add scale text draw.text( (start_x + 5, start_y + 5), f"{scale_mm} mm", fill=(255, 255, 255), font=font ) return pil_img def add_scale_label_v2(pil_img, pixel_sizes, slice_dim): """Add scale label to image.""" draw = ImageDraw.Draw(pil_img) # Get image dimensions - in PIL, size returns (width, height) img_width, img_height = pil_img.size # Define a class with the _get_appropriate_scale method class ScaleCalculator: def _get_appropriate_scale(self, pixel_size, img_size, init_scale=10): scales = [1, 2, 5, 10, 15, 20, 25, 30, 40, 50, 60, 70, 80, 90, 100] scale_pixels_num = int(init_scale / pixel_size) min_pixels = img_size * 0.05 max_pixels = img_size * 0.25 if scale_pixels_num < min_pixels: for scale in scales: if scale > init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) elif scale_pixels_num > max_pixels: for scale in reversed(scales): if scale < init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) return init_scale, scale_pixels_num scale_calculator = ScaleCalculator() # Find which dimension is smaller # In the 2D array: height = first dimension, width = second dimension # In pixel_sizes: [height_scale, width_scale] # In PIL image: img_width = second dimension, img_height = first dimension if img_height < img_width: # Height is the smaller dimension pixel_size_min = pixel_sizes[0] # Height pixel size image_dim_min = img_height else: # Width is the smaller dimension pixel_size_min = pixel_sizes[1] # Width pixel size image_dim_min = img_width # Calculate appropriate scale scale_mm, scale_pixels_min = scale_calculator._get_appropriate_scale( pixel_size_min, image_dim_min, init_scale=10 ) # Calculate scale for the other dimension if img_height < img_width: scale_pixels_height = scale_pixels_min scale_pixels_width = int(scale_mm / pixel_sizes[1]) else: scale_pixels_width = scale_pixels_min scale_pixels_height = int(scale_mm / pixel_sizes[0]) # Position for scale bar (5% from the edge) start_x, start_y = int(img_width * 0.05), int(img_height * 0.05) end_x, end_y = start_x + scale_pixels_width, start_y + scale_pixels_height # Set text font and scale line width default_line_width = 2 default_fontsize = 10 line_width = default_line_width font = ImageFont.load_default().font_variant(size=default_fontsize) # Draw horizontal line draw.line( [(start_x, start_y), (end_x, start_y)], fill=(255, 255, 255), width=line_width, ) # Draw vertical line draw.line( [(start_x, start_y), (start_x, end_y)], fill=(255, 255, 255), width=line_width, ) # Add scale text draw.text( (start_x + 5, start_y + 5), f"{scale_mm} mm", fill=(255, 255, 255), font=font ) return pil_img, scale_mm def add_scale_label_autoGreen(pil_img, pixel_sizes, slice_dim): """Add scale label to image.""" draw = ImageDraw.Draw(pil_img) # Get image dimensions - in PIL, size returns (width, height) img_width, img_height = pil_img.size # Define a class with the _get_appropriate_scale method class ScaleCalculator: def _get_appropriate_scale(self, pixel_size, img_size, init_scale=10): scales = [1, 2, 5, 10, 15, 20, 25, 30, 40, 50, 60, 70, 80, 90, 100] scale_pixels_num = int(init_scale / pixel_size) min_pixels = img_size * 0.05 max_pixels = img_size * 0.25 if scale_pixels_num < min_pixels: for scale in scales: if scale > init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) elif scale_pixels_num > max_pixels: for scale in reversed(scales): if scale < init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) return init_scale, scale_pixels_num scale_calculator = ScaleCalculator() # Find which dimension is smaller # In the 2D array: height = first dimension, width = second dimension # In pixel_sizes: [height_scale, width_scale] # In PIL image: img_width = second dimension, img_height = first dimension if img_height < img_width: # Height is the smaller dimension pixel_size_min = pixel_sizes[0] # Height pixel size image_dim_min = img_height else: # Width is the smaller dimension pixel_size_min = pixel_sizes[1] # Width pixel size image_dim_min = img_width # Calculate appropriate scale scale_mm, scale_pixels_min = scale_calculator._get_appropriate_scale( pixel_size_min, image_dim_min, init_scale=10 ) # Calculate scale for the other dimension if img_height < img_width: scale_pixels_height = scale_pixels_min scale_pixels_width = int(scale_mm / pixel_sizes[1]) else: scale_pixels_width = scale_pixels_min scale_pixels_height = int(scale_mm / pixel_sizes[0]) # Position for scale bar (5% from the edge) start_x, start_y = int(img_width * 0.05), int(img_height * 0.05) end_x, end_y = start_x + scale_pixels_width, start_y + scale_pixels_height # Set text font and scale line width default_line_width = 2 default_fontsize = 10 line_width = default_line_width font = ImageFont.load_default().font_variant(size=default_fontsize) text = f"{scale_mm} mm" _, tmp_text_height = _get_text_dimensions(draw, text, font) if tmp_text_height < img_height * 0.05: fontsize = int(default_fontsize * (img_height * 0.05) / tmp_text_height) font = ImageFont.load_default().font_variant(size=fontsize) if (img_height * 0.05) / tmp_text_height > 5: line_width = int( default_line_width * (img_height * 0.05) / tmp_text_height / 5 ) # Add scale text draw.text((start_x + 5, start_y + 5), text, fill=(0, 255, 0), font=font) # Draw horizontal line draw.line( [(start_x, start_y), (end_x, start_y)], fill=(0, 255, 0), width=line_width, ) # Draw vertical line draw.line( [(start_x, start_y), (start_x, end_y)], fill=(0, 255, 0), width=line_width, ) return pil_img def add_scale_and_orientation_label(pil_img, pixel_sizes, slice_dim): """Add scale bar and orientation labels to image.""" draw = ImageDraw.Draw(pil_img) # Get image dimensions - in PIL, size returns (width, height) img_width, img_height = pil_img.size # Define a class with the _get_appropriate_scale method class ScaleCalculator: def _get_appropriate_scale(self, pixel_size, img_size, init_scale=10): scales = [1, 2, 5, 10, 15, 20, 25, 30, 40, 50, 60, 70, 80, 90, 100] scale_pixels_num = int(init_scale / pixel_size) min_pixels = img_size * 0.05 max_pixels = img_size * 0.25 if scale_pixels_num < min_pixels: for scale in scales: if scale > init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) elif scale_pixels_num > max_pixels: for scale in reversed(scales): if scale < init_scale: return self._get_appropriate_scale(pixel_size, img_size, scale) return init_scale, scale_pixels_num scale_calculator = ScaleCalculator() # Find which dimension is smaller # In the 2D array: height = first dimension, width = second dimension # In pixel_sizes: [height_scale, width_scale] # In PIL image: img_width = second dimension, img_height = first dimension if img_height < img_width: # Height is the smaller dimension pixel_size_min = pixel_sizes[0] # Height pixel size image_dim_min = img_height else: # Width is the smaller dimension pixel_size_min = pixel_sizes[1] # Width pixel size image_dim_min = img_width # Calculate appropriate scale scale_mm, scale_pixels_min = scale_calculator._get_appropriate_scale( pixel_size_min, image_dim_min, init_scale=10 ) # Calculate scale for the other dimension if img_height < img_width: scale_pixels_height = scale_pixels_min scale_pixels_width = int(scale_mm / pixel_sizes[1]) else: scale_pixels_width = scale_pixels_min scale_pixels_height = int(scale_mm / pixel_sizes[0]) # Position for scale bar (5% from the edge) start_x, start_y = int(img_width * 0.05), int(img_height * 0.05) end_x, end_y = start_x + scale_pixels_width, start_y + scale_pixels_height # Set text font and scale line width default_line_width = 2 default_fontsize = 10 line_width = default_line_width font = ImageFont.load_default().font_variant(size=default_fontsize) # Draw horizontal line draw.line( [(start_x, start_y), (end_x, start_y)], fill=(255, 255, 255), width=line_width, ) # Draw vertical line draw.line( [(start_x, start_y), (start_x, end_y)], fill=(255, 255, 255), width=line_width, ) # Add scale text draw.text( (start_x + 5, start_y + 5), f"{scale_mm} mm", fill=(255, 255, 255), font=font ) # Add orientation labels based on slice_dim label_padding = 10 if slice_dim == 0: draw.text((start_x, end_y + 5), "Anterior", fill=(255, 255, 255), font=font) draw.text((end_x + 5, start_y), "Superior", fill=(255, 255, 255), font=font) elif slice_dim == 1: draw.text((start_x, end_y + 5), "Right", fill=(255, 255, 255), font=font) draw.text((end_x + 5, start_y), "Superior", fill=(255, 255, 255), font=font) else: draw.text((start_x, end_y + 5), "Right", fill=(255, 255, 255), font=font) draw.text((end_x + 5, start_y), "Anterior", fill=(255, 255, 255), font=font) return pil_img