[AI特训营第三期]全流程前沿超轻量PPLCNetV2苹果病害识别

#解压数据集!unzip data/data11591/original.zip

#下载PaddleClas，如果不成功可以多下几次!git clone https://gitee.com/PaddlePaddle/PaddleClas.git

!pip install -r PaddleClas/requirements.txt
!pip install paddleclas

#导包import paddleimport osimport cv2import globimport paddle.nn as nnfrom paddle.io import Datasetimport pandas as pdimport paddle.vision.transforms as Timport numpy as npimport jsonimport seaborn as snsimport matplotlib.pyplot as pltfrom PIL import Imagefrom sklearn import preprocessingfrom paddle.vision.transforms import ToTensor

path = 'original'path_imgs = list(glob.glob(path+'/**/*.jpg'))

labels = list(map(lambda x:os.path.split(os.path.split(x)[0])[1], path_imgs))
file_path = pd.Series(path_imgs, name='File_Path').astype(str)
labels = pd.Series(labels, name='Labels')
data = pd.concat([file_path, labels], axis=1)#打乱数据集data = data.sample(frac=1,random_state =2023).reset_index(drop=True)
data.head()

examples = pd.DataFrame([])
unique_labels = data['Labels'].unique()for label in unique_labels:
    examples = pd.concat([examples, data.query(f"Labels == '{label}'").sample(1)])

fig = plt.figure(figsize=(16, 8))for index, row in examples.reset_index().iterrows():
    ax = plt.subplot(4, 3, index + 1)
    image = cv2.imread(row["File_Path"])
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    image = cv2.resize(image, (256, 256), interpolation=cv2.INTER_AREA)
    plt.imshow(image)
    ax.set_title(row["Labels"].title(),fontsize=20)
    ax.axis("off")

fig.tight_layout(pad=0.1)
plt.show()

counts = data.Labels.value_counts()
sns.barplot(x=counts.index, y=counts)
plt.axhline(y=counts.mean(), lw=3, color="#346c9a")
plt.xlabel('Labels')
plt.ylabel('Count')
plt.xticks(rotation=50);

enc=preprocessing.LabelEncoder()
enc=enc.fit(unique_labels)#训练LabelEncoder

#生成标签文件unique_labels_num = enc.transform(unique_labels)
unique_labels_num
label = pd.DataFrame((unique_labels_num,unique_labels))
label = label.transpose()
label.to_csv('label.txt',sep=' ',header=None,index=None)

data['Labels']=enc.transform(data['Labels'])

data.iloc[0:int(0.7*len(data))].to_csv('train.txt',sep=' ',header=None,index=None)
data.iloc[int(0.7*len(data)):].to_csv('test.txt',sep=' ',header=None,index=None)

#注销此代码，训练SwinTransformer# !python PaddleClas/tools/train.py -c SwinTransformer_small_patch4_window7_224.yaml

#训练PPLCNetV2!python PaddleClas/tools/train.py -c PPLCNetV2_base.yaml  -o Arch.pretrained=True -o Arch.use_ssld=True

!python3 PaddleClas/tools/export_model.py \
    -c PPLCNetV2_base.yaml \
    -o Global.pretrained_model=model/lcnetv2/best_model\
    -o Global.save_inference_dir=deploy/models/PPLCNetV2_base_infer

!python3 PaddleClas/tools/eval.py \
    -c SwinTransformer_small_patch4_window7_224.yaml \
    -o Global.pretrained_model=model/swin/best_model \

!python3 PaddleClas/tools/eval.py \
    -c PPLCNetV2_base.yaml \
    -o Global.pretrained_model=model/lcnetv2/best_model\

test = Image.open('original/Brown_Spot/005586.jpg')
test

#模型推理!python3.7 PaddleClas/deploy/python/predict_cls.py -c inference_cls.yaml -o Global.infer_imgs='original/Grey_spot/015951.jpg'