Yolo3的日志分析详解2-主要分析总的损失值、平均损失值等

日志分析：

#!/usr/bin/env python3

# -*- coding: utf-8 -*-

import pandas as pd

import matplotlib.pyplot as plt

import matplotlib.gridspec as gridspec

#最大迭代次数

max_batches=15000

#

minStep = 18

#其中minStep代表1行Loaded: 0.000024 seconds、10行 Region，1行第n次的结果 1+16+1=18

'''

Loaded: 0.000024 seconds

Region 16 Avg IOU: 0.182902, Class: 0.405722, Obj: 0.503862, No Obj: 0.424742, .5R: 0.000000, .75R: 0.000000,  count: 13

Region 23 Avg IOU: 0.105590, Class: 0.662942, Obj: 0.520905, No Obj: 0.509580, .5R: 0.000000, .75R: 0.000000,  count: 12

Region 16 Avg IOU: 0.130522, Class: 0.408217, Obj: 0.476290, No Obj: 0.424609, .5R: 0.000000, .75R: 0.000000,  count: 14

Region 23 Avg IOU: 0.182063, Class: 0.550706, Obj: 0.494642, No Obj: 0.507713, .5R: 0.070175, .75R: 0.000000,  count: 57

Region 16 Avg IOU: 0.200967, Class: 0.390888, Obj: 0.475645, No Obj: 0.427092, .5R: 0.062500, .75R: 0.000000,  count: 16

Region 23 Avg IOU: 0.143319, Class: 0.580808, Obj: 0.531322, No Obj: 0.509306, .5R: 0.064516, .75R: 0.000000,  count: 31

Region 16 Avg IOU: 0.237835, Class: 0.382465, Obj: 0.380643, No Obj: 0.427353, .5R: 0.066667, .75R: 0.000000,  count: 15

Region 23 Avg IOU: 0.106707, Class: 0.574932, Obj: 0.548107, No Obj: 0.510025, .5R: 0.000000, .75R: 0.000000,  count: 37

Region 16 Avg IOU: 0.163751, Class: 0.414830, Obj: 0.416970, No Obj: 0.425206, .5R: 0.000000, .75R: 0.000000,  count: 3

Region 23 Avg IOU: 0.174204, Class: 0.676216, Obj: 0.571296, No Obj: 0.510338, .5R: 0.045455, .75R: 0.000000,  count: 22

Region 16 Avg IOU: 0.191540, Class: 0.441109, Obj: 0.392017, No Obj: 0.425114, .5R: 0.047619, .75R: 0.000000,  count: 21

Region 23 Avg IOU: 0.144743, Class: 0.646313, Obj: 0.482872, No Obj: 0.507978, .5R: 0.000000, .75R: 0.000000,  count: 30

Region 16 Avg IOU: 0.192162, Class: 0.507017, Obj: 0.467949, No Obj: 0.425880, .5R: 0.000000, .75R: 0.000000,  count: 10

Region 23 Avg IOU: 0.129258, Class: 0.672717, Obj: 0.588547, No Obj: 0.509534, .5R: 0.000000, .75R: 0.000000,  count: 36

Region 16 Avg IOU: 0.182297, Class: 0.803474, Obj: 0.596444, No Obj: 0.424545, .5R: 0.000000, .75R: 0.000000,  count: 4

Region 23 Avg IOU: 0.172792, Class: 0.732352, Obj: 0.541576, No Obj: 0.507713, .5R: 0.041667, .75R: 0.000000,  count: 24

1: 756.629517, 756.629517 avg, 0.000000 rate, 1.112322 seconds, 64 images

'''

#在日志里每10个miniStep，即每10次迭代，一次迭代代表1个batch，会出现 2行（分别是Resize 608或者其它尺寸）

n=10

step=minStep*n+2#18x10+2 #2 Resize 608

#x%step==0 or x%step==1

#x%step!=0 and x%step!=1

lines=int(step*max_batches/n)#lines =91000=182*500

result = pd.read_csv('/home/zhangping/Desktop/yolo_log/ranPianTiny2.txt',

                     skiprows=[x for x in range(lines) if (x%step==0 or x%step==1)],

                     error_bad_lines=False,

                     names=['Total Loss', 'Avg loss','rate','seconds','images'])

#只提取以下类似的行

'''

1: 756.629517, 756.629517 avg, 0.000000 rate, 1.112322 seconds, 64 images

第n次迭代：总的损失值，平均损失值，学习率、时间、图片数量

'''

result1=result.iloc[[i for i in range(minStep*max_batches) if (i+1)%18==0]]#

#总的损失值

totalLoss = result1['Total Loss'].str.split(': ').str[1]#<class 'pandas.core.series.Series'>

totalloss = totalLoss.astype(float)

#平均损失值

avgLoss = result1['Avg loss'].str.split(' ').str[1]#replace

avgLoss = avgLoss.astype(float)

#学习率

rate = result1['rate'].str.split(' ').str[1]

rate = rate.astype(float)

#时间

seconds = result1['seconds'].str.split(' ').str[1]

seconds = avgLoss.astype(float)

#图片数量

images = result1['images'].str.split(' ').str[1]

images = images.astype(float)

#

figure = plt.figure()

gs1 = gridspec.GridSpec(3,2)

#画出总的损失值

ax_totalLoss = figure.add_subplot(gs1[0])

ax_totalLoss.set_xlabel('totalLoss')

ax_totalLoss.set_ylabel('totalLoss')

ax_totalLoss.plot(avgLoss.values,label='totalLoss')

#画出平均损失值

ax_avgloss = figure.add_subplot(gs1[1])

ax_avgloss.set_xlabel('avgloss')

ax_avgloss.set_ylabel('avgloss')

ax_avgloss.plot(avgLoss.values,label='avgloss')

#画出学习率

ax_rate = figure.add_subplot(gs1[2])

ax_rate.set_xlabel('rate')

ax_rate.set_ylabel('rate')

ax_rate.plot(rate.values,label='rate')

#画出时间

ax_seconds = figure.add_subplot(gs1[3])

ax_seconds.set_xlabel('seconds')

ax_seconds.set_ylabel('seconds')

ax_seconds.plot(seconds.values,label='seconds')

#画出图片数量

ax_images = figure.add_subplot(gs1[4])

ax_images.set_xlabel('images')

ax_images.set_ylabel('images')

ax_images.plot(images.values,label='images')

plt.show()

总的损失值、平均损失值等图例：

从以上的图例可以看出 总的损失值和平均损失值都在下降~