[AI达人特训营第三期] keggle木薯病叶分类【基于PaddleClas】

文件

train.csv：

sample_submission.csv 给定公开的测试集内容，正确格式化的示例提交。

[train/test]_tfrecords

label_num_to_disease_map

将PaddleClas克隆到本地，并安装PaddleClas依赖。

 # !git clone https://github.com/PaddlePaddle/PaddleClas.git

# pip install --upgrade -r /home/aistudio/PaddleClas/requirements.txt

解压PaddleClas

! unzip /home/aistudio/PaddleClas.zip

# 解压数据集! unzip /home/aistudio/data/data200382/cassava-leaf-disease-classification.zip -d ./dataset/

# 导入依赖import osimport cv2import numpy as npimport pandas as pdimport matplotlib.pyplot as plt

BASE_PATH = '/home/aistudio/dataset/train_images/1004672608.jpg'image = cv2.imread(BASE_PATH)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
plt.imshow(image)
plt.show()
image.shape  # (600, 800, 3)

train_dataset = pd.read_csv('/home/aistudio/dataset/train.csv')
train_dataset.head()

train_dataset.shape #(21397, 2)

plt.figure(figsize=(6, 3))

train_dataset['label'].value_counts().plot(
    kind='bar',
    color='#558354',
    width=0.7)

plt.xlabel('Label', fontsize=12)
plt.ylabel('Count', fontsize=12)
plt.title('Distribution of Labels', fontsize=15)
plt.xticks(rotation=360)
plt.show()

train_image = []
train_id = []for i in train_dataset['image_id']:
    train_image.append(i)for i in train_dataset['label']:
    train_id.append(i)

train_rate, test_rate, val_rate = 0.5, 0.4, 0.1 # 互粉si1 = int(len(train_image)*train_rate)
i2 = int(len(train_image)*(train_rate + test_rate))
 
train_data = train_image[:i1]
train_label = train_id[:i1]

test_data = train_image[i1:i2]
test_label = train_id[i1:i2]

val_data = train_image[i2:]
val_label = train_id[i2:]print(len(train_data),len(train_label))print(len(test_data),len(test_label))print(len(val_data),len(val_label))

for i in range(len(train_data)):    with open("/home/aistudio/dataset/train_list.txt", "a") as f:
        f.write("/home/aistudio/dataset/train_images/" + train_data[i] + " " + str(train_label[i]) + '\n')for i in range(len(test_data)):    with open("/home/aistudio/dataset/test_list.txt", "a") as f:
        f.write("/home/aistudio/dataset/train_images/" + test_data[i] + " " + str(test_label[i]) + '\n')for i in range(len(val_data)):    with open("/home/aistudio/dataset/val_list.txt", "a") as f:
        f.write("/home/aistudio/dataset/train_images/" + val_data[i] + " " + str(val_label[i]) + '\n')

开始训练

%cd ~/PaddleClas
!python3 tools/train.py \
    -c /home/aistudio/PaddleClas/ppcls/configs/quick_start/ResNet50_vd.yaml \
    -o Global.checkpoints="/home/aistudio/PaddleClas/output/ResNet50_vd/epoch_140" \
    -o Arch.pretrained=True \
    -o Global.device=gpu \

%cd ~/PaddleClas
!python3 tools/eval.py \
    -c /home/aistudio/PaddleClas/ppcls/configs/quick_start/ResNet50_vd.yaml \
    -o Global.pretrained_model=/home/aistudio/PaddleClas/output/ResNet50_vd/best_model

%cd ~/PaddleClas
!python3 tools/infer.py \
    -c /home/aistudio/PaddleClas/ppcls/configs/quick_start/ResNet50_vd.yaml \
    -o Global.pretrained_model=/home/aistudio/PaddleClas/output/ResNet50_vd/best_model

%cd ~/PaddleClas
!python3 tools/export_model.py \
    -c /home/aistudio/PaddleClas/ppcls/configs/quick_start/ResNet50_vd.yaml \
    -o Global.pretrained_model=/home/aistudio/PaddleClas/output/ResNet50_vd/best_model \
    -o Global.save_inference_dir=deploy/models/ResNet50_vd_base_infer

# !zip -r PaddleClas.zip /home/aistudio/PaddleClas