2021 iFLYTEK A.I 开发者大赛 碎米识别挑战赛-初赛榜三

#解压数据集!unzip -oq /home/aistudio/data/data105383/碎米识别挑战赛_数据集.zip -d /home/aistudio/work/

# 下载PaddleSeg包，使用PaddleSeg高层API进行训练!pip install paddleseg

# import random# import os# random.seed(2021)# mask_dir  = '/home/aistudio/work/rice/train/labels'# img_dir = '/home/aistudio/work/rice/train/images'# path_list = list()# for img in os.listdir(img_dir):#     img_path = os.path.join(img_dir,img)#     mask_path = os.path.join(mask_dir,img.replace('jpg', 'png'))#     path_list.append((img_path, mask_path))# random.shuffle(path_list)# ratio = 0.8# train_f = open('/home/aistudio/work/rice/train/train.txt','w') # val_f = open('/home/aistudio/work/rice/train/val.txt' ,'w')# for i ,content in enumerate(path_list):#     img, mask = content#     text = img + ' ' + mask + '\n'#     if i < len(path_list) * ratio:#         train_f.write(text)#     else:#         val_f.write(text)# train_f.close()# val_f.close()

import paddleseg.transforms as Tfrom paddleseg.datasets import Dataset

train_transforms = [
    T.RandomHorizontalFlip(0.5),# 水平翻转
    T.RandomVerticalFlip(0.5),# 垂直翻转
    T.RandomDistort(0.6),
    T.RandomScaleAspect(min_scale=0.8,aspect_ratio=0.5),# 随机缩放
    T.Resize(target_size=(1024,1024)),
    T.Normalize()  # 图像标准化]
val_transforms = [
    T.Resize(target_size=(1024,1024)),
    T.Normalize()

]

dataset_root = '/home/aistudio/work/rice/train'train_path  = '/home/aistudio/work/rice/train/train.txt'val_path  = '/home/aistudio/work/rice/train/val.txt'# 构建训练集train_dataset = Dataset(
    dataset_root=dataset_root,
    train_path=train_path,
    transforms=train_transforms,
    num_classes=2,
    mode='train'

                  )#验证集val_dataset = Dataset(
    dataset_root=dataset_root,
    val_path=val_path,
    transforms=val_transforms,
    num_classes=2,
    mode='val'
                  )

import paddlefrom paddleseg.models import UNetfrom paddleseg.models.losses import  MixedLoss,BCELoss,LovaszSoftmaxLoss

base_lr =0.0025 #初始学习率iters = 4000unet_model = UNet(num_classes=2)#使用unet进行训练#自动调整学习率lr =paddle.optimizer.lr.CosineAnnealingDecay(base_lr, T_max=(iters // 3), last_epoch=0.5) #使用余弦退火调整学习率u_optimizer = paddle.optimizer.Adam(lr, parameters=unet_model.parameters())#构建损失函数mixtureLosses = [BCELoss(),LovaszSoftmaxLoss() ] 
mixtureCoef = [0.7,0.3]
losses = {}
losses['types'] = [MixedLoss(mixtureLosses, mixtureCoef)]
losses['coef'] = [1]

#进行训练from paddleseg.core import train
train(
    model = unet_model,
    train_dataset=train_dataset,
    val_dataset=val_dataset,
    optimizer=u_optimizer,
    save_dir='output/RICE_model_1',
    iters=iters,  
    batch_size=2, 
    save_interval=40,
    log_iters=10,
    num_workers=0,
    losses=losses,
    use_vdl=True
    )

from paddleseg.core import evaluatefrom paddleseg.models import UNet#设置模型model = UNet(num_classes=2)#模型路径model_path = 'UNet_0.68997.pdparams'para_state_dict = paddle.load(model_path)
model.set_dict(para_state_dict)
evaluate(model,val_dataset)

%cd ~import randomimport os

test_path = r"work/rice/test/images"test_lst=[]for test in os.listdir(test_path):              
    test_lst.append(test)  

with open('work/rice/test.txt', 'w') as f:    for line in test_lst:
        f.write(line)
        f.write('\n')

import osfrom paddleseg.models import UNetfrom paddleseg.core import predictimport paddleseg.transforms as T
transforms = T.Compose([
    T.Resize(target_size=(1024, 1024)), 
    T.Normalize()
])

model = UNet(num_classes=2)#生成图片列表image_list = []with open('work/rice/test.txt' ,'r') as f:    for line in f.readlines():
        image_list.append(os.path.join('work/rice/test/images/',line.split()[0]))

predict(
        model,        #模型路径文件
        model_path = 'UNet_0.68997.pdparams',
        transforms=transforms,
        image_list=image_list,
        save_dir='result',
    )

!mkdir results#采取的方法是将RGB图转换为0/255import osimport reimport os.pathimport numpy as npfrom PIL import Image#将RGB图转为0/255#Pixels higher than this will be 255. Otherwise 0.image_path = "result/pseudo_color_prediction"results_path = "results"image_arr=[]for image in os.listdir(image_path):             
    image_arr.append(image)                           

for i in range(len(image_arr)):#Load image and convert to greyscale
    img = Image.open(image_path + "/"+image_arr[i])
    img = img.convert("L")
    imgData = np.asarray(img)
    imgData = ((imgData-np.min(imgData))/(np.max(imgData)-np.min(imgData))*255).astype('uint8')
    im = Image.fromarray(imgData)
    im.save(results_path + "/"+image_arr[i])

# 压缩当前路径所有文件，输出zip文件path='results'import zipfile,os
zipName = 'results.zip' #压缩后文件的位置及名称f = zipfile.ZipFile( zipName, 'w', zipfile.ZIP_DEFLATED )for dirpath, dirnames, filenames in os.walk(path):    for filename in filenames:        print(filename)
        f.write(os.path.join(dirpath,filename))
f.close()