常规赛：PALM病理性近视预测 五月份第8名

import pandas as pdimport random

train_excel_file = 'dataset/Train/Classification.xlsx'pd_list=pd.read_excel(train_excel_file)

pd_list_lenght=len(pd_list)# 乱序pd_list=pd_list.sample(frac=1)
offset=int(pd_list_lenght*0.9)
trian_list=pd_list[:offset]
eval_list=pd_list[offset:]
trian_list.to_csv("train_list.txt", index=None, header=None, sep=' ')
eval_list.to_csv("eval_list.txt", index=None, header=None, sep=' ')

# 设置使用0号GPU卡（如无GPU，执行此代码后仍然会使用CPU训练模型）import matplotlib
matplotlib.use('Agg') 
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'import paddlex as pdx

{'transforms.RandomHorizontalFlip":[0.4,0.5,0.6]}
{'transforms.RandomVerticalFlip':[0.3,0.5,0.7]}
{'transforms.RandomRotate':[rotate_range=30,45,60,75]}

from paddlex.cls import transforms
train_transforms = transforms.Compose([

    transforms.RandomCrop(crop_size=1440),
    transforms.RandomHorizontalFlip(0.9),
    transforms.RandomRotate(rotate_range=45),

    transforms.RandomVerticalFlip(0.2),
    transforms.Normalize()
])
eval_transforms = transforms.Compose([
    transforms.ResizeByShort(short_size=1440),
    transforms.CenterCrop(crop_size=1440),
    transforms.Normalize()
])

from paddlex.cls import transforms
train_transforms = transforms.Compose([

    transforms.RandomCrop(crop_size=224),
    transforms.RandomHorizontalFlip(0.6),
    transforms.RandomRotate(rotate_range=75),

    transforms.RandomVerticalFlip(0.6),
    transforms.Normalize()
])
eval_transforms = transforms.Compose([
    transforms.ResizeByShort(short_size=256),
    transforms.CenterCrop(crop_size=224),
    transforms.Normalize()
])

train_dataset = pdx.datasets.ImageNet(
    data_dir='dataset/Train/fundus_image',
    file_list='train_list.txt',
    label_list='labels.txt',
    transforms=train_transforms,
    shuffle=True)
eval_dataset = pdx.datasets.ImageNet(
    data_dir='dataset/Train/fundus_image',
    file_list='eval_list.txt',
    label_list='labels.txt',
    transforms=eval_transforms)

# model = pdx.cls.ResNet50_vd_ssld(num_classes=2)# model = pdx.cls.ResNet101_vd_ssld(num_classes=2)model = pdx.cls.MobileNetV3_large_ssld(num_classes=2)
model.train(num_epochs=25,
            train_dataset=train_dataset,
            train_batch_size=8,
            eval_dataset=eval_dataset,
            lr_decay_epochs=[10, 15, 20],
            save_interval_epochs=5,
            learning_rate=0.03,
            save_dir='output/ResNet50_vd_ssld7',
            use_vdl=True)

# 预测数据集val_listval_list=[]for i in range(1,401,1):
    filename='T'+ str(i).zfill(4)+'.jpg'
    print(filename)
    val_list.append(filename+'\n')with open('val_list.txt','w') as f:
    f.writelines(val_list)
    
val_list=[]with open('val_list.txt', 'r') as f:    for line in f:
        line='dataset/PALM-Testing400-Images/'+line
        val_list.append(line.split('\n')[0])        # print(line.split('\n')[0])# print(val_list)

# 批量预测import numpy as np
model = pdx.load_model('output/ResNet50_vd_ssld7/best_model')# model = pdx.load_model('best_model')results = []for file in val_list:
    result = model.predict(file,topk=2)
    results.append(result)# result = model.batch_predict(val_list)
    print("Predict Result:", result)

# 结果列pd_B=[]for item in results:    # print(item)
    if item[0]['category_id']==1:
        pd_B.append(item[0]['score'])    else:
        pd_B.append(1-item[0]['score'])

# 文件名列pd_A=[]with open('val_list.txt', 'r') as f:    for line in f:
        pd_A.append(line.split('\n')[0])        # print(line.split('\n')[0])

# 构造pandas的DataFrameimport pandas as pd
df= pd.DataFrame({'FileName': pd_A, 'PM Risk':pd_B})

# 保存为提交文件df.to_csv("Classification_Results062101.csv", index=None)