图像语义分割EDA模板

# 解压数据集!unzip -oq /home/aistudio/data/data68698/智能车数据集.zip -d /home/aistudio/data

%cd data

%run make_list.py

import osimport cv2import numpy as npfrom tqdm import tqdmimport matplotlib.pyplot as pltimport osfrom collections import Counterimport pandas as pdfrom pandas.core.frame import DataFramefrom PIL import Imageimport random

%matplotlib inline

images/xxx1.jpg (xx1.png) annotations/xxx1.png
images/xxx2.jpg (xx2.png) annotations/xxx2.png
……

labelA
labelB
……

#amount of classerCLASSES_NUM = 15#find image in folder dirdef findImages(dir,topdown=True):
    im_list = []    if not os.path.exists(dir):        print("Path for {} not exist!".format(dir))        raise
    else:        for root, dirs, files in os.walk(dir, topdown):            for fl in files:
                im_list.append(fl)    return im_list# 每一类别所包含的图像数量images_count = [0]*CLASSES_NUM# 每一类别的像素数目class_pixels_count = [0]*CLASSES_NUM# 每一类别对应的图像的总像素数目image_pixels_count = [0]*CLASSES_NUM

image_folder = './mask_4000'im_list = findImages(image_folder) 

# 创建一个统计全部图片像素标签信息的DataFramesum_df = pd.DataFrame()for im in tqdm(im_list):    # 读取文件名
    # print(im)
    # 加载图片
    cv_img = cv2.imread(os.path.join(image_folder, im), cv2.IMREAD_UNCHANGED)
    size_img = cv_img.shape
    colors = set([])
    agg = []    for i in range(size_img[0]):        for j in range(size_img[1]):            # (i,j)位置像素点label
            p_value = cv_img.item(i,j)            # 检查像素点类别数是否正确
            if not p_value < CLASSES_NUM: # check
                print(p_value)            else:                # 给指定类别计数+1
                class_pixels_count[p_value] = class_pixels_count[p_value] + 1
                # 把像素点塞到set集合里
                colors.add(p_value)                # print(p_value)
                agg.append(p_value)
    im_size = size_img[0]*size_img[1]
    df = pd.DataFrame.from_dict(Counter(agg), orient='index')
    df = df.T
    df['filename'] = im
    df['sum_pixels'] = im_size    # print(df)
    sum_df = sum_df.append(df, ignore_index=True)

sum_df = sum_df.fillna(0)

# 分析列重排序column_names = ['filename', 'sum_pixels']
classes = [i for i in range(CLASSES_NUM)]
column_names.extend(classes)    
sum_df = sum_df.loc[:,column_names]

# 生成各类别目标占比数据for i in range(CLASSES_NUM):
    sum_df[str(i) + '_ratio'] = sum_df[i] /  sum_df['sum_pixels']

sum_df.head()

# 保存分析数据sum_df.to_csv('../sum_df.csv', sep=',', header=True, index=True)

a = sum_df.loc[:,classes].apply(lambda x:x.sum(),axis=0)

plt.figure(figsize=(12,12)) #调节图形大小labels = [str(i) for i in range(CLASSES_NUM)] #定义标签patches,text1,text2 = plt.pie(a,
                      labels=labels,
                      autopct = '%3.2f%%', #数值保留固定小数位
                      shadow = False, #无阴影设置
                      startangle =90, #逆时针起始角度设置
                      pctdistance = 0.6) #数值距圆心半径倍数距离#patches饼图的返回值，texts1饼图外label的文本，texts2饼图内部的文本# x，y轴刻度设置一致，保证饼图为圆形plt.axis('equal')
plt.show()

mask_pt = '/home/aistudio/data/mask_4000'count =  [0 for _ in range(16)]for m in tqdm(os.listdir(mask_pt)):
    pt = os.path.join(mask_pt, m)
    mask = cv2.imread(pt, 0)
    size = mask.shape[0] * mask.shape[1]    for i in range(16):
        a = mask == i
        ratio_i = np.sum(a.astype(np.float)) / size
        count[i] += ratio_i

sum_ = np.sum(count[1:])
ratios = [v/sum_ for v in count[1:]]

for i in range(0, len(ratios)):    print('-[INFO] Label {}: {:.4f}'.format(i+1, ratios[i]))

plt.figure(figsize=(24,24)) #调节图形大小labels = [str(i) for i in range(1, 16)] #定义标签patches,text1,text2 = plt.pie(ratios,
                      labels=labels,
                      autopct = '%3.2f%%', #数值保留固定小数位
                      shadow = False, #无阴影设置
                      startangle =90, #逆时针起始角度设置
                      pctdistance = 0.6) #数值距圆心半径倍数距离#patches饼图的返回值，texts1饼图外label的文本，texts2饼图内部的文本# x，y轴刻度设置一致，保证饼图为圆形plt.axis('equal')
plt.show()

image_pt = '/home/aistudio/data/image_4000/'pseudo_color_pt = '/home/aistudio/data/color_4000/'

!mkdir color_4000

!python ../gray2pseudo_color.py mask_4000 color_4000

def plot_image_examples(df, rows=3, cols=3, title='Image examples'):
    fig, axs = plt.subplots(rows, cols, figsize=(16,16))    if title=='Image examples':        for row in range(rows):            for col in range(cols):
                idx = np.random.randint(len(df), size=1)[0]
                name = df.iloc[idx]["filename"]
                img = Image.open(image_pt + str(name))
                axs[row, col].imshow(img)                
                axs[row, col].axis('off')    else:        for row in range(rows):            for col in range(cols):
                idx = np.random.randint(len(df), size=1)[0]
                name = df.iloc[idx]["filename"]
                img = Image.open(pseudo_color_pt + str(name))
                axs[row, col].imshow(img)                
                axs[row, col].axis('off')
    plt.suptitle(title)

sum_df = pd.read_csv('../sum_df.csv',index_col='Unnamed: 0')

sum_df.head()

less_spikes_ids = sum_df[sum_df['1']>0][sum_df['1']<500].sort_values(by='1',ascending=False).filename
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)])

plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)],title='Mask examples')

less_spikes_ids = sum_df[sum_df['2']>0][sum_df['2']<500].sort_values(by='2',ascending=False).filename
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)])
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)],title='Mask examples')

less_spikes_ids = sum_df[sum_df['1']>max(sum_df['1'])*0.9].sort_values(by='1',ascending=False).filename
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)])
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)],title='Mask examples')

less_spikes_ids = sum_df[sum_df['2']>max(sum_df['2'])*0.9].sort_values(by='2',ascending=False).filename
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)])
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)],title='Mask examples')

sum_df[sum_df['1']>0][sum_df['2']>0][sum_df['3']>0].sort_values(by=str(i),ascending=False).filename
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)])
plot_image_examples(sum_df[sum_df.filename.isin(less_spikes_ids)],title='Mask examples')

class paddleseg.models.losses.OhemCrossEntropyLoss(
                thresh = 0.7,
                min_kept = 10000,
                ignore_index = 255
)

class paddleseg.models.losses.OhemEdgeAttentionLoss(
                edge_threshold = 0.8,
                thresh = 0.7,
                min_kept = 5000,
                ignore_index = 255
)