### 基于COCO数据集的物体大小分类标准处理YOLO格式标签文件 为了实现这一目标，可以按照以下方式构建 Python 脚本： #### 1. COCO 物体尺寸定义 根据 COCO 数据集的标准，物体被分为三类：小、中、大。具体划分如下[^1]: - 小 (small): 面积小于 \(32^2\) 的像素 - 中 (medium): 面积介于 \([32^2, 96^2)\) 的像素之间 - 大 (large): 面积大于等于 \(96^2\) 的像素 #### 2. YOLO 标签解析逻辑 YOLO 格式的标签文件通常每行表示一个边界框，格式为 `class_id center_x center_y width height`，这些数值都是相对于图片宽高的比例。 要计算实际面积，需要先读取对应的图像宽度和高度信息，再通过下面公式得到绝对值： \[ area = image\_width * label\_width * image\_height * label\_height\] #### 3. 统计代码实现 以下是完整的 Python 实现方案： ```python import os def calculate_area(image_width, image_height, yolo_bbox): """Calculate the absolute area from normalized YOLO bbox.""" _, x_center, y_center, box_w, box_h = map(float, yolo_bbox.split()) abs_box_w = image_width * box_w abs_box_h = image_height * box_h return abs_box_w * abs_box_h def classify_object_size(area): """Classify object size based on COCO standard.""" small_threshold = pow(32, 2) medium_threshold = pow(96, 2) if area < small_threshold: return 'small' elif small_threshold <= area < medium_threshold: return 'medium' else: return 'large' def count_objects(label_file_path, image_width=640, image_height=480): """ Count objects by their sizes. :param label_file_path: Path to .txt file containing YOLO labels :param image_width: Width of corresponding image in pixels :param image_height: Height of corresponding image in pixels :return: Dictionary with counts per category and overall total """ stats = {'total': 0, 'small': 0, 'medium': 0, 'large': 0} try: with open(label_file_path, "r") as f: lines = f.readlines() for line in lines: area = calculate_area(image_width, image_height, line.strip()) obj_class = classify_object_size(area) stats[obj_class] += 1 stats['total'] += 1 except Exception as e: print(f"Error processing {label_file_path}: ", str(e)) return stats if __name__ == "__main__": folder_path = "./labels" all_stats = {} # Iterate over each txt file within specified directory for filename in os.listdir(folder_path): if not filename.endswith(".txt"): continue filepath = os.path.join(folder_path, filename) result = count_objects(filepath) key_name = filename.replace('.txt', '') all_stats[key_name] = result # Print results for name, stat in all_stats.items(): print(f"\nFile: {name}") for k, v in stat.items(): print(f"{k.capitalize()} objects:", v) ``` 此脚本会遍历指定目录下的所有 `.txt` 文件，并针对每一个文件执行上述操作来统计不同尺寸的对象数目。