【PaddleHub模型贡献】一行代码实现驾驶员状态识别

    'c0': 'normal driving',

    'c1': 'texting-right',
    
    'c2': 'talking on the phone-right',
    
    'c3': 'texting-left',
    
    'c4': 'talking on the phone-left',
    
    'c5': 'operating the radio',
    
    'c6': 'drinking',
    
    'c7': 'reaching behind',
    
    'c8': 'hair and makeup',
    
    'c9': 'talking to passenger'

# 解压数据集（注意要根据环境修改路径——!unzip /home/aistudio/data/data35503//imgs.zip -d /home/aistudio/work/imgs
!cp /home/aistudio/data/data35503/lbls.csv /home/aistudio/work/

!pip install paddlex -i https://mirror.baidu.com/pypi/simple

import os# 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）os.environ['CUDA_VISIBLE_DEVICES'] = '0'

# jupyter中使用paddlex需要设置matplotlibimport matplotlibimport os
os.chdir('/home/aistudio/work/')
matplotlib.use('Agg') 
import paddlex as pdx

import paddlehub as hubimport paddle.fluid as fluidimport numpy as np

base = './imgs/'datas = []for i in range(10):
    c_base = base+'train/c{}/'.format(i)    for im in os.listdir(c_base):
        pt = os.path.join('train/c{}/'.format(i), im)
        line = '{} {}'.format(pt, i)        # print(line)
        datas.append(line)

np.random.seed(10)
np.random.shuffle(datas)

total_num = len(datas)
train_num = int(0.8*total_num)
test_num = int(0.1*total_num)
valid_num = total_num - train_num - test_numprint('train:', train_num)print('valid:', valid_num)print('test:', test_num)with open(base+'train_list.txt', 'w') as f:    for v in datas[:train_num]:
        f.write(v+'\n')with open(base+'test_list.txt', 'w') as f:    for v in datas[-test_num:]:
        f.write(v+'\n')with open(base+'val_list.txt', 'w') as f:    for v in datas[train_num:-test_num]:
        f.write(v+'\n')with open(base+'labels.txt', 'w') as f:    for i in range(10):
        f.write('ch{}\n'.format(i))

from paddlex.cls import transforms
train_transforms = transforms.Compose([
    transforms.ResizeByShort(short_size=256),
    transforms.RandomCrop(crop_size=224),
    transforms.RandomDistort(),
    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='imgs',
    file_list='imgs/train_list.txt',
    label_list='imgs/labels.txt',
    transforms=train_transforms,
    shuffle=True)
eval_dataset = pdx.datasets.ImageNet(
    data_dir='imgs',
    file_list='imgs/val_list.txt',
    label_list='imgs/labels.txt',
    transforms=eval_transforms)

num_classes = len(train_dataset.labels)print(num_classes)

# 定义并训练模型model = pdx.cls.MobileNetV3_small_ssld(num_classes=num_classes)
model.train(num_epochs=20,
            train_dataset=train_dataset,
            train_batch_size=32,
            log_interval_steps=20,
            eval_dataset=eval_dataset,
            lr_decay_epochs=[1],
            save_interval_epochs=1,
            learning_rate=0.01,
            save_dir='output/mobilenetv3')

2021-03-12 22:53:09 [INFO]	[TRAIN] Epoch=20/20, Step=500/560, loss=0.117403, acc1=0.96875, acc5=1.0, lr=0.001, time_each_step=1.7s, eta=0:2:40
2021-03-12 22:53:43 [INFO]	[TRAIN] Epoch=20/20, Step=520/560, loss=0.057465, acc1=1.0, acc5=1.0, lr=0.001, time_each_step=1.68s, eta=0:2:5
2021-03-12 22:54:12 [INFO]	[TRAIN] Epoch=20/20, Step=540/560, loss=0.521278, acc1=0.84375, acc5=0.96875, lr=0.001, time_each_step=1.44s, eta=0:1:27
2021-03-12 22:54:37 [INFO]	[TRAIN] Epoch=20/20, Step=560/560, loss=0.20346, acc1=0.9375, acc5=1.0, lr=0.001, time_each_step=1.29s, eta=0:0:58
2021-03-12 22:54:37 [INFO]	[TRAIN] Epoch 20 finished, loss=0.297043, acc1=0.89721, acc5=0.986719, lr=0.001 .
2021-03-12 22:54:37 [INFO]	Start to evaluating(total_samples=2243, total_steps=71)...
100%|██████████| 71/71 [00:57<00:00,  1.23it/s]
2021-03-12 22:55:35 [INFO]	[EVAL] Finished, Epoch=20, acc1=0.988854, acc5=1.0 .
2021-03-12 22:55:35 [INFO]	Model saved in output/mobilenetv3/best_model.
2021-03-12 22:55:36 [INFO]	Model saved in output/mobilenetv3/epoch_20.
2021-03-12 22:55:36 [INFO]	Current evaluated best model in eval_dataset is epoch_20, acc1=0.9888542131074454

!paddlex --export_inference --model_dir=output/mobilenetv3/best_model --save_dir=inference_model

save_dir = 'output/mobilenetv3/best_model'model = pdx.load_model(save_dir)
model.evaluate(eval_dataset, batch_size=1, epoch_id=None, return_details=False)

!hub convert --model_dir work/inference_model \
              --module_name DriverStatusRecognition \
              --module_version 1.0.0 \
              --output_dir outputs

!gzip -dfq outputs/DriverStatusRecognition.tar.gz
!tar -xf outputs/DriverStatusRecognition.tar

!hub install DriverStatusRecognition

[2021-03-13 12:09:07,593] [    INFO] - Successfully uninstalled DriverStatusRecognition
[2021-03-13 12:09:07,785] [    INFO] - Successfully installed DriverStatusRecognition-1.0.0

import cv2import paddlehub as hub

module = hub.Module(directory='DriverStatusRecognition') # 一行代码实现模型调用images = [cv2.imread('work/imgs/test/img_1622.jpg'), cv2.imread('work/imgs/test/img_14165.jpg'), cv2.imread('work/imgs/test/img_47183.jpg')]
results = module.predict(images=images)for result in results:    print(result)