[map, loss, pr, F1 drawing in YOLO series result] use matplotlib to draw the result file generated after training according to v5, v8, v7

1. Rendering

2. Understand the result content

2.1 YOLOv7的result.txt

• Reference link: YOLOv7 result analysis, txt file content

0/299      14.7G         0.07522         0.009375          0.02266          0.1073                    58            640           0.0002958    0.1458   0.0002676     4.469e-05          0.1005          0.01098       0.02545
训练轮数   GPU消耗     train/box_loss   train/obj_loss    train/cls_loss  train/total_loss     本轮中数据中目标数量   输入图片大小         P          R         mAP@.5     mAP@.5:.95        val/Box_loss  val/obj_loss   val/cls_loss

2.2 YOLOv5 | YOLOv8 result.csv (each column has the same meaning)

The meaning of each column is the same , probably because the same author wrote the code, so the saved content of result.csv has not changed

YOLOv5

YOLOv8

3. Code (only map and loss)

• What needs to be changed : only the content in, each line represents andresult_dict模型名称该模型训练得到的result文件地址
• Both v5 and v8 are csvin format, and v7 is txtin format
• Key points to pay attention to : In the following code 关于下标的注释位置, the subscripts are all 0incremented from the beginning to the right
• plt.savefig("mAP50.png", dpi=600)# dpi can be set to 300/600/900, which means saving as a higher-definition vector image
• data = pd.read_csv(res_path, usecols=[6]).values.ravel()Indicates that after reading csvthe file, only take the 6value of the subscripted column values, and ravel()flatten the value into a one-dimensional array (if you want to draw result.csv/result.txtother values, just change the corresponding subscript directly. For example, if you want to draw train/box_loss, Then 绘制map50change 6 to 1, and change 10 to 2)
• plt.plot(x, data, label=modelname, linewidth='1')# Line thickness is set to 1

import matplotlib.pyplot as plt
import pandas as pd
import numpy as np

if __name__ == '__main__':
    # 列出待获取数据内容的文件位置
    # v5、v8都是csv格式的，v7是txt格式的
    result_dict = {
    
    
        'YOLOv5m': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov5m\results.csv',
        'YOLOv7': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7\results.txt',
        'YOLOv7-tiny': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7-tiny\results.txt',
        'YOLOv7-tiny-large': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7-tiny-large\results.txt',
        'YOLOv7-tiny-PConv': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7-tiny-PConv\results.txt',
        'YOLOv7-tiny-GhostNetv2': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7-tiny-GhostNetv2\results.txt',
        # 'YOLOv8s': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov8s\results.csv',
        'YOLOv7-tiny-large-GhostNetv2': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7-tiny-large-GhostNetv2\results.txt',
        'YOLOv7-GhostNetv2': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7-GhostNetv2\results.txt',
        'YOLOv7-FasterNet': r'C:\Users\dadandan\Desktop\workspace_temp\NeimengA_train_resultcsv\yolov7-FasterNet\results.txt'
    }

    # 绘制map50
    for modelname in result_dict:
        res_path = result_dict[modelname]
        ext = res_path.split('.')[-1]
        if ext == 'csv':
            data = pd.read_csv(res_path, usecols=[6]).values.ravel()    # 6是指map50的下标（每行从0开始向右数）
        else:   # 文件后缀是txt
            with open(res_path, 'r') as f:
                datalist = f.readlines()
                data = []
                for d in datalist:
                    data.append(float(d.strip().split()[10]))   # 10是指map50的下标（每行从0开始向右数）
                data = np.array(data)
        x = range(len(data))
        plt.plot(x, data, label=modelname, linewidth='1')   # 线条粗细设为1

    # 添加x轴和y轴标签
    plt.xlabel('Epochs')
    plt.ylabel('[email protected]')
    plt.legend()
    plt.grid()
    # 显示图像
    plt.savefig("mAP50.png", dpi=600)   # dpi可设为300/600/900，表示存为更高清的矢量图
    plt.show()


    # 绘制map50-95
    for modelname in result_dict:
        res_path = result_dict[modelname]
        ext = res_path.split('.')[-1]
        if ext == 'csv':
            data = pd.read_csv(res_path, usecols=[7]).values.ravel()    # 7是指map50-95的下标（每行从0开始向右数）
        else:
            with open(res_path, 'r') as f:
                datalist = f.readlines()
                data = []
                for d in datalist:
                    data.append(float(d.strip().split()[11]))   # 11是指map50-95的下标（每行从0开始向右数）
                data = np.array(data)
        x = range(len(data))
        plt.plot(x, data, label=modelname, linewidth='1')

    # 添加x轴和y轴标签
    plt.xlabel('Epochs')
    plt.ylabel('[email protected]:0.95')
    plt.legend()
    plt.grid()
    # 显示图像
    plt.savefig("mAP50-95.png", dpi=600)
    plt.show()

    # 绘制训练的总loss
    for modelname in result_dict:
        res_path = result_dict[modelname]
        ext = res_path.split('.')[-1]
        if ext == 'csv':
            box_loss = pd.read_csv(res_path, usecols=[1]).values.ravel()
            obj_loss = pd.read_csv(res_path, usecols=[2]).values.ravel()
            cls_loss = pd.read_csv(res_path, usecols=[3]).values.ravel()
            data = np.round(box_loss + obj_loss + cls_loss, 5)    # 3个loss相加并且保留小数点后5位（与v7一致）

        else:
            with open(res_path, 'r') as f:
                datalist = f.readlines()
                data = []
                for d in datalist:
                    data.append(float(d.strip().split()[5]))
                data = np.array(data)
        x = range(len(data))
        plt.plot(x, data, label=modelname, linewidth='1')

    # 添加x轴和y轴标签
    plt.xlabel('Epochs')
    plt.ylabel('Loss')
    plt.legend()
    plt.grid()
    # 显示图像
    plt.savefig("loss.png", dpi=600)
    plt.show()

4. For drawing PR and F1, check this article

• [Map drawing in YOLO series result] Use matplotlib to draw the result file generated after training according to v5, v8, and v7