基于PaddleX2.0-PP-LCNet模型的咖啡豆质检分类

!unzip -oq /home/aistudio/data/data150459/dataset3.zip -d work/

!pip install paddlex

* 划分数据集
* 生成标签

!paddlex --split_dataset --format ImageNet --dataset_dir work/dataset3/ --val_value 0.2

2022-06-04 16:27:51 [INFO]	Dataset split starts...
2022-06-04 16:27:51 [INFO]	Dataset split done.
2022-06-04 16:27:51 [INFO]	Train samples: 709
2022-06-04 16:27:51 [INFO]	Eval samples: 177
2022-06-04 16:27:51 [INFO]	Test samples: 0
2022-06-04 16:27:51 [INFO]	Split files saved in work/dataset3/

* PP-LCNet模型组网
* 调整训练参数

import paddlex as pdxfrom paddlex import transforms as T# 定义训练和验证时的transforms# API说明：https://github.com/PaddlePaddle/PaddleX/blob/develop/docs/apis/transforms/transforms.mdtrain_transforms = T.Compose(
    [T.RandomCrop(crop_size=224), T.RandomHorizontalFlip(), T.Normalize()])

eval_transforms = T.Compose([
    T.ResizeByShort(short_size=256), T.CenterCrop(crop_size=224), T.Normalize()
])# 定义训练和验证所用的数据集# API说明：https://github.com/PaddlePaddle/PaddleX/blob/develop/docs/apis/datasets.mdtrain_dataset = pdx.datasets.ImageNet(
    data_dir='work/dataset3/',
    file_list='work/dataset3/train_list.txt',
    label_list='work/dataset3/labels.txt',
    transforms=train_transforms,
    shuffle=True)

eval_dataset = pdx.datasets.ImageNet(
    data_dir='work/dataset3/',
    file_list='work/dataset3/val_list.txt',
    label_list='work/dataset3/labels.txt',
    transforms=eval_transforms)# 初始化模型，并进行训练# 可使用VisualDL查看训练指标，参考https://github.com/PaddlePaddle/PaddleX/blob/develop/docs/visualdl.mdnum_classes = len(train_dataset.labels)
model = pdx.cls.PPLCNet(num_classes=num_classes, scale=1)# API说明：https://github.com/PaddlePaddle/PaddleX/blob/develop/docs/apis/models/classification.md# 各参数介绍与调整说明：https://github.com/PaddlePaddle/PaddleX/tree/develop/docs/parameters.mdmodel.train(
    num_epochs=100,
    pretrain_weights='IMAGENET',
    train_dataset=train_dataset,
    train_batch_size=64,
    eval_dataset=eval_dataset,
    lr_decay_epochs=[4, 6, 8],
    learning_rate=0.1,
    save_dir='output/pplcnet',
    log_interval_steps=10,
    label_smoothing=.1,
    use_vdl=True)

2022-06-04 16:37:16 [INFO]	Model saved in output/pplcnet/epoch_99.
2022-06-04 16:37:17 [INFO]	[TRAIN] Epoch=100/100, Step=1/11, loss=0.204105, acc1=1.000000, acc2=1.000000, lr=0.000025, time_each_step=0.31s, eta=0:0:3
2022-06-04 16:37:19 [INFO]	[TRAIN] Epoch=100/100, Step=11/11, loss=0.259414, acc1=0.953125, acc2=1.000000, lr=0.000000, time_each_step=0.24s, eta=0:0:0
2022-06-04 16:37:19 [INFO]	[TRAIN] Epoch 100 finished, loss=0.22206952, acc1=0.984375, acc2=1.0 .
2022-06-04 16:37:20 [INFO]	Start to evaluate(total_samples=177, total_steps=3)...
2022-06-04 16:37:21 [INFO]	[EVAL] Finished, Epoch=100, acc1=0.982781, acc2=1.000000 .
2022-06-04 16:37:21 [INFO]	Current evaluated best model on eval_dataset is epoch_64, acc1=0.9879889488220215
2022-06-04 16:37:21 [INFO]	Model saved in output/pplcnet/epoch_100.

* 最终训练结果acc1=0.98.8，训练的效果非常不错，可以达到应用级的效果

import paddlex as pdx
test_jpg = 'IMG_20200618_124521_1.jpg'model = pdx.load_model('output/pplcnet/best_model')
result = model.predict(test_jpg)print("Predict Result: ", result)

2022-06-04 20:22:09 [INFO]	Model[PPLCNet] loaded.Predict Result:  [{'category_id': 1, 'category': 'Debris_Shell', 'score': 0.93589294}]