PythonWorkGroup-0/ver3.py

from fastai.data.transforms import get_image_files, parent_label, RandomSplitter, Normalize
from fastai.learner import load_learner
from fastai.metrics import error_rate
from pathlib import Path
from fastai.data.block import CategoryBlock, DataBlock
from fastai.vision.all import *
from fastai.vision.augment import Resize, aug_transforms
from fastai.vision.core import imagenet_stats
from fastai.vision.data import ImageBlock
from fastai.vision.learner import cnn_learner, vision_learner
from torchvision.models import resnet34
from PIL import Image


def open_image(image_path):
    img = Image.open(image_path)
    img_cvt = img.resize((460,460))
    return img_cvt

def predict_image(image_path):
    # 加载模型
    model = load_learner('G:\\Users\\15819\\Desktop\\model01.pkl')
    # 读取图片并转换为Tensor
    img = open_image(image_path)  # 读取指定路径（image_path）下的图像文件
    # 进行预测
    pred_class, pred_idx, outputs = model.predict(img)
    # 获取置信度
    # 检查输出张量的维度
    if outputs.dim() == 0:
        confidence = float(outputs)
    else:
        confidence = float(outputs[pred_idx])

    return pred_class, confidence

def train():
    data_path = Path('G:\\Users\\15819\\Desktop\\TrainSet')
    blocks = (ImageBlock, CategoryBlock)
    batch_size = 32
    dls = DataBlock(
        blocks=blocks,
        get_items=get_image_files,
        splitter=RandomSplitter(),
        get_y=parent_label,
        item_tfms=Resize(460),
        batch_tfms=[*aug_transforms(size=224, min_scale=0.75), Normalize.from_stats(*imagenet_stats)]
    ).dataloaders(data_path, num_workers=4, bs=batch_size)

    model = vision_learner(dls, resnet34, metrics=error_rate)
    model.fine_tune(5, freeze_epochs=3) #5 - 训练的轮次， 3 - 冻结的轮次
    model.export('G:\\Users\\15819\\Desktop\\model01.pkl') # M20.02.pkl

def main():
    image_path = 'G:\\Users\\15819\\Desktop\\TrainSet\\SmallCar\\京M88888.jpg'
    pred_class, confidence = predict_image(image_path)
    print(f"图片类别： {pred_class}, 置信度： {confidence}")


if __name__ == '__main__':
    train()