Python脚本实现统计与可视化COCO数据集标注信息

pip install matplotlib

import json
from collections import Counter
import matplotlib.pyplot as plt
def analyze_coco_json(file_path):
    print(f"正在加载并分析文件: {file_path} ...\n")
    with open(file_path, 'r', encoding='utf-8') as f:
        coco_data = json.load(f)
    # 1. 基础信息提取
    images = coco_data.get('images', [])
    annotations = coco_data.get('annotations', [])
    categories_info = coco_data.get('categories', [])
    num_images = len(images)
    num_annotations = len(annotations)
    categories = {cat['id']: cat['name'] for cat in categories_info}
    print("=== 1. 数据集基础统计 ===")
    print(f"图片总数: {num_images}")
    print(f"标注框总数: {num_annotations}")
    print(f"类别映射字典: {categories}\n")
    # 2. 统计各类别标注数量
    cat_counts = Counter([ann['category_id'] for ann in annotations])
    cat_distribution = {categories.get(k, f"Unknown ({k})"): v for k, v in cat_counts.items()}
    print("=== 2. 类别分布情况 ===")
    # 按数量从多到少排序
    sorted_cats = dict(sorted(cat_distribution.items(), key=lambda item: item[1], reverse=True))
    for cat, count in sorted_cats.items():
        print(f"  [{cat}]: {count} 个标注")
    # 3. 统计图片分辨率
    resolutions = Counter([(img['width'], img['height']) for img in images])
    print("\n=== 3. 图片分辨率分布 ===")
    for res, count in sorted(resolutions.items(), key=lambda x: x[1], reverse=True):
        print(f"  {res[0]}x{res[1]}: {count} 张图片")
    # 4. 绘制类别分布柱状图并保存
    plt.figure(figsize=(8, 5))
    plt.bar(sorted_cats.keys(), sorted_cats.values(), color='#4C72B0')
    plt.title('Annotation Count per Category in COCO Dataset')
    plt.xlabel('Categories')
    plt.ylabel('Number of Annotations (BBoxes)')
    plt.xticks(rotation=45)
    # 在柱状图上显示具体数值
    for i, (cat, count) in enumerate(sorted_cats.items()):
        plt.text(i, count + sum(sorted_cats.values())*0.01, str(count), ha='center')
    plt.tight_layout()
    plot_filename = 'category_distribution.png'
    plt.savefig(plot_filename, dpi=300)
    print(f"\n[成功] 类别分布可视化图表已保存为: '{plot_filename}'")
if __name__ == "__main__":
    # 请确保将此处文件名替换为您实际的文件路径
    analyze_coco_json('_annotations.coco.json')

python analyze_coco.py

plt.rcParams['font.sans-serif'] = ['SimHei']  # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False   # 用来正常显示负号