YOLO系列训练生成的exp进行处理找出mAP值

weixin_45734021

已于 2024-09-25 20:21:33 修改

阅读量413

点赞数 4

文章标签： YOLO

于 2024-09-25 20:18:03 首次发布

本文链接：https://blog.csdn.net/weixin_45734021/article/details/142531900

版权

YOLOv8或者YOLOv10训练生成的exp文件进行遍历，找出results.csv文件中metrics/mAP50(B)、metrics/mAP50-95(B)值和args.yaml中的mode文件，将他们保存到一个excel文件中。

YOLO训练train.py代码如下：
在这里插入图片描述
模型训练好后生成exp文件如下：

生成excel文件完整代码如下：

import os
import pandas as pd
import yaml

def find_max_map50_and_model(root_dir):
    results = []

    for subdir, _, _ in os.walk(root_dir):
        # 获取每个子文件夹中的所有文件夹
        subfolders = [f.path for f in os.scandir(subdir) if f.is_dir()]

        for folder in subfolders:
            # 检查每个文件夹中是否包含results.csv和args.yaml文件
            results_csv_path = os.path.join(folder, 'results.csv')
            args_yaml_path = os.path.join(folder, 'args.yaml')

            if os.path.isfile(results_csv_path) and os.path.isfile(args_yaml_path):
                # 读取 results.csv 文件
                df = pd.read_csv(results_csv_path)
                print(df.columns)
                if '       metrics/mAP50(B)' in df.columns and '    metrics/mAP50-95(B)' in df.columns:
                    max_value = df['       metrics/mAP50(B)'].max()
                    max_row = df[df['       metrics/mAP50(B)'] == max_value]
                    max_map50_95_value = max_row['    metrics/mAP50-95(B)'].values[0]
                    precision = max_row['   metrics/precision(B)'].values[0]
                    recall = max_row['      metrics/recall(B)'].values[0]

                    # 读取 args.yaml 文件
                    with open(args_yaml_path, 'r') as f:
                        args = yaml.safe_load(f)
                        model_info = args.get('model', 'N/A')
                        model_info = model_info.split('/')[-1].split(".")[0]

                    results.append({
                        'folder': folder,
                        'metrics/mAP50(B)': max_value,
                        'metrics/mAP50-95(B)': max_map50_95_value,
                        'Model': model_info,
                        "recall":recall,
                        "precision":precision
                    })

    return results
output_file = r'results_summary.xlsx'  #生成xlsx文件
root_directory = r'runs/train'      #训练保存的exp问你就
results = find_max_map50_and_model(root_directory)

# 将结果转换为DataFrame
df_results = pd.DataFrame(results, columns=['Model','precision','recall','parameters','GFLOPS' ,'metrics/mAP50(B)', 'metrics/mAP50-95(B)', 'folder'])
# df_results = pd.DataFrame(results)

# 将DataFrame保存为Excel文件
df_results.to_excel(output_file, index=False)
print(f"Results have been saved to {output_file}")

# 读取保存的Excel文件
df_saved = pd.read_excel(output_file)

# 找到metrics/mAP50(B)的最大值
max_map50_value = df_saved['metrics/mAP50(B)'].max()
max_row = df_saved[df_saved['metrics/mAP50(B)'] == max_map50_value]

print("Row with the maximum metrics/mAP50(B):")
print(max_map50_value)

主要是改动下面这两个：

output_file = r'results_summary.xlsx'  #生成xlsx文件
root_directory = r'runs/train'      #训练保存的exp问你就

在这里插入图片描述