import copy import ast from PIL import Image, ImageDraw, ImageFont, ImageColor additional_colors = [colorname for (colorname, colorcode) in ImageColor.colormap.items()] def plot_bounding_boxes(im, bounding_boxes, input_width, input_height): """ Plots bounding boxes on an image with markers for each a name, using PIL, normalized coordinates, and different colors. Args: img_path: The path to the image file. bounding_boxes: A list of bounding boxes containing the name of the object and their positions in normalized [y1 x1 y2 x2] format. """ # Load the image img = im width, height = img.size # print(img.size) # Create a drawing object draw = ImageDraw.Draw(img) # Define a list of colors colors = [ 'red', 'green', 'blue', 'yellow', 'orange', 'pink', 'purple', 'brown', 'gray', 'beige', 'turquoise', 'cyan', 'magenta', 'lime', 'navy', 'maroon', 'teal', 'olive', 'coral', 'lavender', 'violet', 'gold', 'silver', ] + additional_colors if bounding_boxes is None: # Display the image return # font = ImageFont.truetype("NotoSansCJK-Regular.ttc", size=14) # Iterate over the bounding boxes for i, bounding_box in enumerate(bounding_boxes): # Select a color from the list if len(bounding_box["bbox_2d"]) != 4: continue color = colors[i % len(colors)] # if bounding_box["bbox_2d"][0] <= 1.: # input_height, input_width = 1., 1. # Convert normalized coordinates to absolute coordinates abs_x1 = int(bounding_box["bbox_2d"][0]/input_width * width) abs_y1 = int(bounding_box["bbox_2d"][1]/input_height * height) abs_x2 = int(bounding_box["bbox_2d"][2]/input_width * width) abs_y2 = int(bounding_box["bbox_2d"][3]/input_height * height) # print(bounding_box["bbox_2d"], abs_x1, abs_y1, abs_x2, abs_y2) if abs_x1 > abs_x2: abs_x1, abs_x2 = abs_x2, abs_x1 if abs_y1 > abs_y2: abs_y1, abs_y2 = abs_y2, abs_y1 # print("(abs_x1, abs_y1), (abs_x2, abs_y2):", (abs_x1, abs_y1), (abs_x2, abs_y2), bounding_box["bbox_2d"], input_width, width, input_height, height) # print(draw) # Draw the bounding box draw.rectangle( ((abs_x1, abs_y1), (abs_x2, abs_y2)), outline=color, width=4 ) # Draw the text if "label" in bounding_box: draw.text((abs_x1 + 8, abs_y1 + 6), bounding_box["label"], fill=color) # font=font # Display the image def plot_movement(im, data, input_width, input_height): img = im width, height = img.size draw = ImageDraw.Draw(img) colors = ['red', 'blue'] # Assume red for starting, blue for ending line_width = 4 if data is None: # Display the image return # print("decode json points: ", data, len(data)) for line in data: # Extract starting and ending points start_point = line.get("start_point_2d", None) end_point = line.get("end_point_2d", None) if start_point is None or end_point is None: # print("Starting or ending location not found. ", line) return # if start_point[0] <= 1.: # input_height, input_width = 1., 1. # Converting coordinates abs_x_start = int(start_point[0]) / input_width * width abs_y_start = int(start_point[1]) / input_height * height abs_x_end = int(end_point[0]) / input_width * width abs_y_end = int(end_point[1]) / input_height * height # Draw an arrow from the starting point to the ending point draw.line((abs_x_start, abs_y_start, abs_x_end, abs_y_end), fill='black', width=line_width) # Draw and annotate start and end points for i, point in enumerate([start_point, end_point]): color = colors[i % len(colors)] abs_x = int(point[0]) / input_width * width abs_y = int(point[1]) / input_height * height radius = 4 # label = data[i]["label"] draw.ellipse([(abs_x - radius, abs_y - radius), (abs_x + radius, abs_y + radius)], fill=color) # draw.text((abs_x + 8, abs_y + 6), label, fill=color) def safe_eval(item): try: # 使用 ast.literal_eval 代替 eval，以安全地评估字符串表达式 return ast.literal_eval(item) except (ValueError, SyntaxError) as e: # 如果解析失败，记录错误信息并返回 None # print(f"Failed to evaluate item: {item}. Error: {e}") return None def parse_json(json_output): # Parsing out the markdown fencing json_output_list = [] lines = json_output.splitlines() for i, line in enumerate(lines): # print(i, line, line.strip()=="```json") if line.strip() == "```json": tmp = "\n".join(lines[i+1:]) # Remove everything before "```json" if "```" in tmp: tmp = tmp.split("```")[0] # Remove everything after the closing "```" else: tmp = tmp.split("")[0].strip() json_output_list.append(tmp) return json_output_list def parse_bbox_and_movement(response): parsed_list = parse_json(response) parsed_list = [safe_eval(item) for item in parsed_list] parsed_list = [item for item in parsed_list if item is not None] # filter bbox_list, movement_list = [], [] for item_list in parsed_list: for item in item_list: if "bbox_2d" in item and "label" in item: bbox_list.append(item) elif "start_point_2d" in item and "end_point_2d" in item and "label" in item: movement_list.append(item) # for item in bbox_list: # # item = eval(item) # print(type(item), item) # print("\n########################") # for item in movement_list: # print(type(item), item) return bbox_list, movement_list def merge_bbox_list(bbox_list_origin, bbox_list_new, image_index_new): """ 合并两个边界框列表，确保每个标签的边界框唯一，优先使用 bbox_list_new 中的数据。 :param bbox_list_origin: 原始边界框列表 :param bbox_list_new: 新的边界框列表 :return: 合并后的边界框列表 """ image_index = -1 if len(bbox_list_new): # 构建新标签字典，用于快速查找 new_label_dict = {bbox['label']: bbox for bbox in bbox_list_new} merged_bbox_list = [] try: image_index = [bb['index']-1 for bb in bbox_list_new] except Exception as e: print('bbox_list_new:', bbox_list_new) raise e assert len(set(image_index)) == 1, f"bbox_list_new: {bbox_list_new}" image_index = image_index[0] if image_index not in bbox_list_origin: bbox_list_origin[image_index] = [] # 遍历原始列表，按需替换或保留边界框 for bbox in bbox_list_origin[image_index]: label = bbox['label'] if label in new_label_dict: # 优先使用新列表中的边界框，并标记为已处理 merged_bbox_list.append(new_label_dict.pop(label)) else: # 保留原始列表中的边界框 merged_bbox_list.append(bbox) # 将新列表中剩余的边界框添加到结果中 merged_bbox_list.extend(new_label_dict.values()) bbox_list_origin[image_index_new] = copy.deepcopy(merged_bbox_list) return image_index, bbox_list_origin def merge_movement_list(movement_list_origin, movement_list_new, image_index_new): """ 合并两个移动方向列表，确保每个标签的移动方向唯一，优先使用 movement_list_new 中的数据。 :param movement_list_origin: 原始移动方向列表 :param movement_list_new: 新的移动方向列表 :return: 合并后的移动方向列表 """ image_index = -1 if len(movement_list_new): # 构建新标签字典，用于快速查找 new_label_dict = {movement['label']: movement for movement in movement_list_new} merged_movement_list = [] image_index = [mv['index'] - 1 for mv in movement_list_new] assert len(set(image_index)) == 1 image_index = image_index[0] if image_index not in movement_list_origin: movement_list_origin[image_index] = [] # 遍历原始列表，按需替换或保留移动方向 for movement in movement_list_origin[image_index]: label = movement['label'] if label in new_label_dict: # 优先使用新列表中的移动方向，并标记为已处理 merged_movement_list.append(new_label_dict.pop(label)) else: # 保留原始列表中的移动方向 merged_movement_list.append(movement) # 将新列表中剩余的移动方向添加到结果中 merged_movement_list.extend(new_label_dict.values()) movement_list_origin[image_index_new] = copy.deepcopy(merged_movement_list) return image_index, movement_list_origin def merge_bbox_movement(bbox_list_origin, movement_list_origin, bbox_list_new, movement_list_new, image_index_new): idx1, merged_bbox_list = merge_bbox_list(bbox_list_origin, bbox_list_new, image_index_new) idx2, merged_movement_list = merge_movement_list(movement_list_origin, movement_list_new, image_index_new) if idx1 >= 0 and idx2 == -1: merged_movement_list[image_index_new] = copy.deepcopy(merged_movement_list[idx1]) # else: # merged_movement_list[image_index_new] = [] if idx2 >= 0 and idx1 == -1: merged_bbox_list[image_index_new] = copy.deepcopy(merged_bbox_list[idx2]) # else: # merged_bbox_list[image_index_new] = [] if image_index_new not in merged_bbox_list: merged_bbox_list[image_index_new] = [] if image_index_new not in merged_movement_list: merged_movement_list[image_index_new] = [] return max([idx1, idx2]), merged_bbox_list, merged_movement_list