
                        # 载入项目文件夹
import sys
sys.path.append(r"D:\\Workspace\\MyProjects\\ButterflyClassification")
#################################################################################


import os
import time
import json
from utils.draw import draw_process
from utils.logger import logger
from utils.optimizer import learning_rate_setting, optimizer_setting
from utils.reader import train_reader, trainval_reader, val_reader
from config import *
from eval import *
import paddle
from paddle.static import InputSpec

# 初始配置变量
total_epoch = train_parameters['total_epoch']

# 初始化绘图列表
all_train_iters = []
all_train_losses = []
all_train_accs_top1 = []
all_train_accs_top5 = []
all_test_losses = []
all_test_iters = []
all_test_accs_top1 = []
all_test_accs_top5 = []

def train(model):  
    # 初始化临时变量
    num_batch = 0
    best_result = 0
    best_result_id = 0
    elapsed = 0


    # 根据config文件设置训练数据来源
    if train_parameters['training_data'] == 'trainval':
        data_reader = trainval_reader
    elif train_parameters['training_data'] == 'train':
        data_reader = train_reader

    for epoch in range(1, total_epoch+1):
        for batch_id, (image, label) in enumerate(data_reader()):
            num_batch += 1
            
            label = paddle.unsqueeze(label, axis=1)                        
            loss, acc = model.train_batch([image], [label])
       
            if num_batch % train_parameters['log_interval'] == 0: # 每10个batch显示一次日志，适合大数据集  
                avg_loss = loss[0][0]
                acc_top1 = acc[0][0]
                acc_top5 = acc[0][1]
                
                elapsed_step = time.perf_counter() - elapsed - start
                elapsed = time.perf_counter() - start
                logger.info('Epoch:{}/{}, batch:{}, train_loss:[{:.5f}], acc_top1:[{:.5f}], acc_top5:[{:.5f}]({:.2f}s)'
                            .format(epoch, total_epoch, num_batch, loss[0][0], acc[0][0], acc[0][1], elapsed_step))

                # 记录训练过程，用于可视化训练过程中的loss和accuracy
                all_train_iters.append(num_batch)
                all_train_losses.append(avg_loss)
                all_train_accs_top1.append(acc_top1)
                all_train_accs_top5.append(acc_top5)

        # 每隔一定周期进行一次测试
        if epoch % train_parameters['eval_interval'] == 0 or epoch == total_epoch:
            # 模型校验
            avg_loss, avg_acc_top1, avg_acc_top5 = eval(model, val_reader())            
            logger.info('[validation] Epoch:{}/{}, val_loss:[{:.5f}], val_top1:[{:.5f}], val_top5:[{:.5f}]'.format(epoch, total_epoch, avg_loss, avg_acc_top1, avg_acc_top5))

            # 记录测试过程，用于可视化训练过程中的loss和accuracy
            all_test_iters.append(epoch)
            all_test_losses.append(avg_loss)
            all_test_accs_top1.append(avg_acc_top1)            
            all_test_accs_top5.append(avg_acc_top5)
            

            # 将性能最好的模型保存为final模型
            if avg_acc_top1 > best_result:
                best_result = avg_acc_top1
                best_result_id = epoch
                 
                # finetune model 用于调优和恢复训练
                model.save(os.path.join(checkpoint_path, model_name + '_final')) 
                # inference model 用于部署和预测 
                model.save(os.path.join(final_models_path, model_name + '_final'), training=False)
                logger.info('已保存当前测试模型(epoch={})为最优模型:{}_final'.format(best_result_id, model_name)) 
            logger.info('最优top1测试精度:{:.5f} (epoch={})'.format(best_result, best_result_id))

            # 根据需要决定是否需要将每次测试的模型都进行保存(if needed)，保存模型需要耗费一定的运算时间和大量的存储资源
            # 建议在训练大型模型时，开启该选项，方便训练中断时能够及时恢复训练
            # 训练小型模型（训练时间短）时，可以关闭该选项，以进一步提高训练速度
            if train_parameters['checkpointed']:
                model.save(os.path.join(checkpoint_path, model_name + '_' + str(epoch)))

    logger.info('训练完成，最终性能accuracy={:.5f}(epoch={}), 总耗时{:.2f}s, 已将其保存为：{}_final'.format(best_result, best_result_id, time.perf_counter() - start, model_name))


if __name__ == '__main__':
    # 将此次训练的超参数进行保存
    data = json.dumps(train_parameters, indent=4, ensure_ascii=False, sort_keys=False, separators=(',', ':'))   # 格式化字典格式的参数列表
    logger.info(data)
    # 启动训练过程
    logger.info('训练参数保存完毕，使用{}模型, 训练{}数据, 训练集{}, 启动训练...'.format(train_parameters['architecture'],train_parameters['dataset_name'],train_parameters['training_data']))
    logger.info('当前模型目录为：{}'.format(model_name + '_' + train_parameters['starting_time']))

    # 设置输入样本的维度
    input_spec = InputSpec(shape=[None] + train_parameters['input_size'], dtype='float32', name='image')
    label_spec = InputSpec(shape=[None, 1], dtype='int64', name='label')
  
    # 载入官方标准模型，若不存在则会自动进行下载，pretrained=True|False控制是否使用Imagenet预训练参数
    network = paddle.vision.models.resnet50(num_classes=train_parameters['class_dim'], pretrained=train_parameters['pretrained'])  
    model = paddle.Model(network, input_spec, label_spec) 
    logger.info('模型参数信息：')
    logger.info(model.summary()) # 是否显示神经网络的具体信息
    
    if train_parameters['checkpoint_train'] == True:
        model.load(checkpoint_load_model)
        logger.info('载入{}中断模型和参数完毕，开始从checkpoint恢复训练'.format(train_parameters['architecture']))
        logger.info('checkpoint模型:{}'.format(checkpoint_load_model))
    else:
        if train_parameters['pretrained'] == False:
            logger.info('载入{}模型完毕，从初始状态开始训练'.format(train_parameters['architecture']))
        elif train_parameters['pretrained_model'] == 'API':
            logger.info('载入Imagenet-{}预训练模型完毕，开始微调训练(fine-tune)'.format(train_parameters['architecture']))
        else:
            model.load(project_pretrained_model)
            logger.info('载入自定义预训练{}模型完毕，开始微调训练(fine-tune)'.format(train_parameters['architecture']))
            logger.info('预训练模型:{}'.format(project_pretrained_model))
          
    # 设置学习率、优化器、损失函数和评价指标
    lr = learning_rate_setting()
    optimizer = optimizer_setting(model, lr)    
    model.prepare(optimizer,
                  paddle.nn.CrossEntropyLoss(),
                  paddle.metric.Accuracy(topk=(1,5)))   

    # 启动训练过程
    start = time.perf_counter()
    train(model)
    logger.info('训练完毕，结果路径{}.'.format(result_root_path))

    # 输出训练过程图
    logger.info('Done.')
    draw_process("Training Process", 'Train Loss', 'Train Accuracy(top1)', all_train_iters, all_train_losses, all_train_accs_top1, 'train')
    draw_process("Validation Results", 'Validation Loss', 'Validation Accuracy(top1)', all_test_iters, all_test_losses, all_test_accs_top1, 'val')

2021-03-06 19:40:59,114 - INFO: {
    "dataset_name":"Butterfly",
    "architecture":"Mobilenetv2_without_pretrained",
    "training_data":"train",
    "starting_time":"202103061920",
    "input_size":[
        3,
        224,
        224
    ],
    "mean_value":[
        0.485,
        0.456,
        0.406
    ],
    "std_value":[
        0.229,
        0.224,
        0.225
    ],
    "num_trainval":490,
    "num_train":423,
    "num_val":67,
    "num_test":129,
    "class_dim":7,
    "label_dict":{
        "0":"admiral",
        "1":"black_swallowtail",
        "2":"machaon",
        "3":"monarch_closed",
        "4":"monarch_open",
        "5":"peacock",
        "6":"zebra"
    },
    "total_epoch":2,
    "batch_size":64,
    "log_interval":1,
    "eval_interval":1,
    "checkpointed":false,
    "checkpoint_train":false,
    "checkpoint_model":"Butterfly_Mobilenetv2_pretrained_final",
    "checkpoint_time":"202102182058",
    "pretrained":true,
    "pretrained_model":"API",
    "dataset_root_path":"D:\\Workspace\\ExpDatasets",
    "result_root_path":"D:\\Workspace\\ExpResults",
    "project_result_path":"D:\\Workspace\\MyProjects\\ButterflyClassification\\results",
    "useGPU":true,
    "learning_strategy":{
        "optimizer_strategy":"Momentum",
        "learning_rate_strategy":"CosineAnnealingDecay",
        "learning_rate":0.01,
        "momentum":0.9,
        "Piecewise_boundaries":[
            60,
            80,
            90
        ],
        "Piecewise_values":[
            0.01,
            0.001,
            0.0001,
            1e-05
        ],
        "Exponential_gamma":0.9,
        "Polynomial_decay_steps":10,
        "verbose":false
    },
    "augmentation_strategy":{
        "withAugmentation":true,
        "augmentation_prob":0.5,
        "rotate_angle":15,
        "Hflip_prob":0.5,
        "brightness":0.4,
        "contrast":0.4,
        "saturation":0.4,
        "hue":0.4
    }
}
2021-03-06 19:40:59,115 - INFO: 训练参数保存完毕，使用Mobilenetv2_without_pretrained模型, 训练Butterfly数据, 训练集train, 启动训练...
2021-03-06 19:40:59,115 - INFO: 当前模型目录为：Butterfly_Mobilenetv2_without_pretrained_202103061920
2021-03-06 19:40:59,190 - INFO: unique_endpoints {''}
2021-03-06 19:40:59,192 - INFO: File C:\Users\Administrator/.cache/paddle/hapi/weights\resnet50.pdparams md5 checking...
2021-03-06 19:40:59,497 - INFO: Found C:\Users\Administrator/.cache/paddle/hapi/weights\resnet50.pdparams
C:\Users\Administrator\anaconda3\lib\site-packages\paddle\fluid\dygraph\layers.py:1263: UserWarning: Skip loading for fc.weight. fc.weight receives a shape [2048, 1000], but the expected shape is [2048, 7].
  warnings.warn(("Skip loading for {}. ".format(key) + str(err)))
C:\Users\Administrator\anaconda3\lib\site-packages\paddle\fluid\dygraph\layers.py:1263: UserWarning: Skip loading for fc.bias. fc.bias receives a shape [1000], but the expected shape is [7].
  warnings.warn(("Skip loading for {}. ".format(key) + str(err)))
2021-03-06 19:41:00,184 - INFO: 模型参数信息：
2021-03-06 19:41:00,218 - INFO: {'total_params': 23575495, 'trainable_params': 23469255}
2021-03-06 19:41:00,219 - INFO: 载入Imagenet-Mobilenetv2_without_pretrained预训练模型完毕，开始微调训练(fine-tune)

-------------------------------------------------------------------------------
   Layer (type)         Input Shape          Output Shape         Param #    
===============================================================================
    Conv2D-262       [[1, 3, 224, 224]]   [1, 64, 112, 112]        9,408     
  BatchNorm2D-262   [[1, 64, 112, 112]]   [1, 64, 112, 112]         256      
      ReLU-18       [[1, 64, 112, 112]]   [1, 64, 112, 112]          0       
    MaxPool2D-2     [[1, 64, 112, 112]]    [1, 64, 56, 56]           0       
    Conv2D-264       [[1, 64, 56, 56]]     [1, 64, 56, 56]         4,096     
  BatchNorm2D-264    [[1, 64, 56, 56]]     [1, 64, 56, 56]          256      
      ReLU-19        [[1, 256, 56, 56]]    [1, 256, 56, 56]          0       
    Conv2D-265       [[1, 64, 56, 56]]     [1, 64, 56, 56]        36,864     
  BatchNorm2D-265    [[1, 64, 56, 56]]     [1, 64, 56, 56]          256      
    Conv2D-266       [[1, 64, 56, 56]]     [1, 256, 56, 56]       16,384     
  BatchNorm2D-266    [[1, 256, 56, 56]]    [1, 256, 56, 56]        1,024     
    Conv2D-263       [[1, 64, 56, 56]]     [1, 256, 56, 56]       16,384     
  BatchNorm2D-263    [[1, 256, 56, 56]]    [1, 256, 56, 56]        1,024     
BottleneckBlock-17   [[1, 64, 56, 56]]     [1, 256, 56, 56]          0       
    Conv2D-267       [[1, 256, 56, 56]]    [1, 64, 56, 56]        16,384     
  BatchNorm2D-267    [[1, 64, 56, 56]]     [1, 64, 56, 56]          256      
      ReLU-20        [[1, 256, 56, 56]]    [1, 256, 56, 56]          0       
    Conv2D-268       [[1, 64, 56, 56]]     [1, 64, 56, 56]        36,864     
  BatchNorm2D-268    [[1, 64, 56, 56]]     [1, 64, 56, 56]          256      
    Conv2D-269       [[1, 64, 56, 56]]     [1, 256, 56, 56]       16,384     
  BatchNorm2D-269    [[1, 256, 56, 56]]    [1, 256, 56, 56]        1,024     
BottleneckBlock-18   [[1, 256, 56, 56]]    [1, 256, 56, 56]          0       
    Conv2D-270       [[1, 256, 56, 56]]    [1, 64, 56, 56]        16,384     
  BatchNorm2D-270    [[1, 64, 56, 56]]     [1, 64, 56, 56]          256      
      ReLU-21        [[1, 256, 56, 56]]    [1, 256, 56, 56]          0       
    Conv2D-271       [[1, 64, 56, 56]]     [1, 64, 56, 56]        36,864     
  BatchNorm2D-271    [[1, 64, 56, 56]]     [1, 64, 56, 56]          256      
    Conv2D-272       [[1, 64, 56, 56]]     [1, 256, 56, 56]       16,384     
  BatchNorm2D-272    [[1, 256, 56, 56]]    [1, 256, 56, 56]        1,024     
BottleneckBlock-19   [[1, 256, 56, 56]]    [1, 256, 56, 56]          0       
    Conv2D-274       [[1, 256, 56, 56]]    [1, 128, 56, 56]       32,768     
  BatchNorm2D-274    [[1, 128, 56, 56]]    [1, 128, 56, 56]         512      
      ReLU-22        [[1, 512, 28, 28]]    [1, 512, 28, 28]          0       
    Conv2D-275       [[1, 128, 56, 56]]    [1, 128, 28, 28]       147,456    
  BatchNorm2D-275    [[1, 128, 28, 28]]    [1, 128, 28, 28]         512      
    Conv2D-276       [[1, 128, 28, 28]]    [1, 512, 28, 28]       65,536     
  BatchNorm2D-276    [[1, 512, 28, 28]]    [1, 512, 28, 28]        2,048     
    Conv2D-273       [[1, 256, 56, 56]]    [1, 512, 28, 28]       131,072    
  BatchNorm2D-273    [[1, 512, 28, 28]]    [1, 512, 28, 28]        2,048     
BottleneckBlock-20   [[1, 256, 56, 56]]    [1, 512, 28, 28]          0       
    Conv2D-277       [[1, 512, 28, 28]]    [1, 128, 28, 28]       65,536     
  BatchNorm2D-277    [[1, 128, 28, 28]]    [1, 128, 28, 28]         512      
      ReLU-23        [[1, 512, 28, 28]]    [1, 512, 28, 28]          0       
    Conv2D-278       [[1, 128, 28, 28]]    [1, 128, 28, 28]       147,456    
  BatchNorm2D-278    [[1, 128, 28, 28]]    [1, 128, 28, 28]         512      
    Conv2D-279       [[1, 128, 28, 28]]    [1, 512, 28, 28]       65,536     
  BatchNorm2D-279    [[1, 512, 28, 28]]    [1, 512, 28, 28]        2,048     
BottleneckBlock-21   [[1, 512, 28, 28]]    [1, 512, 28, 28]          0       
    Conv2D-280       [[1, 512, 28, 28]]    [1, 128, 28, 28]       65,536     
  BatchNorm2D-280    [[1, 128, 28, 28]]    [1, 128, 28, 28]         512      
      ReLU-24        [[1, 512, 28, 28]]    [1, 512, 28, 28]          0       
    Conv2D-281       [[1, 128, 28, 28]]    [1, 128, 28, 28]       147,456    
  BatchNorm2D-281    [[1, 128, 28, 28]]    [1, 128, 28, 28]         512      
    Conv2D-282       [[1, 128, 28, 28]]    [1, 512, 28, 28]       65,536     
  BatchNorm2D-282    [[1, 512, 28, 28]]    [1, 512, 28, 28]        2,048     
BottleneckBlock-22   [[1, 512, 28, 28]]    [1, 512, 28, 28]          0       
    Conv2D-283       [[1, 512, 28, 28]]    [1, 128, 28, 28]       65,536     
  BatchNorm2D-283    [[1, 128, 28, 28]]    [1, 128, 28, 28]         512      
      ReLU-25        [[1, 512, 28, 28]]    [1, 512, 28, 28]          0       
    Conv2D-284       [[1, 128, 28, 28]]    [1, 128, 28, 28]       147,456    
  BatchNorm2D-284    [[1, 128, 28, 28]]    [1, 128, 28, 28]         512      
    Conv2D-285       [[1, 128, 28, 28]]    [1, 512, 28, 28]       65,536     
  BatchNorm2D-285    [[1, 512, 28, 28]]    [1, 512, 28, 28]        2,048     
BottleneckBlock-23   [[1, 512, 28, 28]]    [1, 512, 28, 28]          0       
    Conv2D-287       [[1, 512, 28, 28]]    [1, 256, 28, 28]       131,072    
  BatchNorm2D-287    [[1, 256, 28, 28]]    [1, 256, 28, 28]        1,024     
      ReLU-26       [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-288       [[1, 256, 28, 28]]    [1, 256, 14, 14]       589,824    
  BatchNorm2D-288    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
    Conv2D-289       [[1, 256, 14, 14]]   [1, 1024, 14, 14]       262,144    
  BatchNorm2D-289   [[1, 1024, 14, 14]]   [1, 1024, 14, 14]        4,096     
    Conv2D-286       [[1, 512, 28, 28]]   [1, 1024, 14, 14]       524,288    
  BatchNorm2D-286   [[1, 1024, 14, 14]]   [1, 1024, 14, 14]        4,096     
BottleneckBlock-24   [[1, 512, 28, 28]]   [1, 1024, 14, 14]          0       
    Conv2D-290      [[1, 1024, 14, 14]]    [1, 256, 14, 14]       262,144    
  BatchNorm2D-290    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
      ReLU-27       [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-291       [[1, 256, 14, 14]]    [1, 256, 14, 14]       589,824    
  BatchNorm2D-291    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
    Conv2D-292       [[1, 256, 14, 14]]   [1, 1024, 14, 14]       262,144    
  BatchNorm2D-292   [[1, 1024, 14, 14]]   [1, 1024, 14, 14]        4,096     
BottleneckBlock-25  [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-293      [[1, 1024, 14, 14]]    [1, 256, 14, 14]       262,144    
  BatchNorm2D-293    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
      ReLU-28       [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-294       [[1, 256, 14, 14]]    [1, 256, 14, 14]       589,824    
  BatchNorm2D-294    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
    Conv2D-295       [[1, 256, 14, 14]]   [1, 1024, 14, 14]       262,144    
  BatchNorm2D-295   [[1, 1024, 14, 14]]   [1, 1024, 14, 14]        4,096     
BottleneckBlock-26  [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-296      [[1, 1024, 14, 14]]    [1, 256, 14, 14]       262,144    
  BatchNorm2D-296    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
      ReLU-29       [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-297       [[1, 256, 14, 14]]    [1, 256, 14, 14]       589,824    
  BatchNorm2D-297    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
    Conv2D-298       [[1, 256, 14, 14]]   [1, 1024, 14, 14]       262,144    
  BatchNorm2D-298   [[1, 1024, 14, 14]]   [1, 1024, 14, 14]        4,096     
BottleneckBlock-27  [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-299      [[1, 1024, 14, 14]]    [1, 256, 14, 14]       262,144    
  BatchNorm2D-299    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
      ReLU-30       [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-300       [[1, 256, 14, 14]]    [1, 256, 14, 14]       589,824    
  BatchNorm2D-300    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
    Conv2D-301       [[1, 256, 14, 14]]   [1, 1024, 14, 14]       262,144    
  BatchNorm2D-301   [[1, 1024, 14, 14]]   [1, 1024, 14, 14]        4,096     
BottleneckBlock-28  [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-302      [[1, 1024, 14, 14]]    [1, 256, 14, 14]       262,144    
  BatchNorm2D-302    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
      ReLU-31       [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-303       [[1, 256, 14, 14]]    [1, 256, 14, 14]       589,824    
  BatchNorm2D-303    [[1, 256, 14, 14]]    [1, 256, 14, 14]        1,024     
    Conv2D-304       [[1, 256, 14, 14]]   [1, 1024, 14, 14]       262,144    
  BatchNorm2D-304   [[1, 1024, 14, 14]]   [1, 1024, 14, 14]        4,096     
BottleneckBlock-29  [[1, 1024, 14, 14]]   [1, 1024, 14, 14]          0       
    Conv2D-306      [[1, 1024, 14, 14]]    [1, 512, 14, 14]       524,288    
  BatchNorm2D-306    [[1, 512, 14, 14]]    [1, 512, 14, 14]        2,048     
      ReLU-32        [[1, 2048, 7, 7]]     [1, 2048, 7, 7]           0       
    Conv2D-307       [[1, 512, 14, 14]]     [1, 512, 7, 7]       2,359,296   
  BatchNorm2D-307     [[1, 512, 7, 7]]      [1, 512, 7, 7]         2,048     
    Conv2D-308        [[1, 512, 7, 7]]     [1, 2048, 7, 7]       1,048,576   
  BatchNorm2D-308    [[1, 2048, 7, 7]]     [1, 2048, 7, 7]         8,192     
    Conv2D-305      [[1, 1024, 14, 14]]    [1, 2048, 7, 7]       2,097,152   
  BatchNorm2D-305    [[1, 2048, 7, 7]]     [1, 2048, 7, 7]         8,192     
BottleneckBlock-30  [[1, 1024, 14, 14]]    [1, 2048, 7, 7]           0       
    Conv2D-309       [[1, 2048, 7, 7]]      [1, 512, 7, 7]       1,048,576   
  BatchNorm2D-309     [[1, 512, 7, 7]]      [1, 512, 7, 7]         2,048     
      ReLU-33        [[1, 2048, 7, 7]]     [1, 2048, 7, 7]           0       
    Conv2D-310        [[1, 512, 7, 7]]      [1, 512, 7, 7]       2,359,296   
  BatchNorm2D-310     [[1, 512, 7, 7]]      [1, 512, 7, 7]         2,048     
    Conv2D-311        [[1, 512, 7, 7]]     [1, 2048, 7, 7]       1,048,576   
  BatchNorm2D-311    [[1, 2048, 7, 7]]     [1, 2048, 7, 7]         8,192     
BottleneckBlock-31   [[1, 2048, 7, 7]]     [1, 2048, 7, 7]           0       
    Conv2D-312       [[1, 2048, 7, 7]]      [1, 512, 7, 7]       1,048,576   
  BatchNorm2D-312     [[1, 512, 7, 7]]      [1, 512, 7, 7]         2,048     
      ReLU-34        [[1, 2048, 7, 7]]     [1, 2048, 7, 7]           0       
    Conv2D-313        [[1, 512, 7, 7]]      [1, 512, 7, 7]       2,359,296   
  BatchNorm2D-313     [[1, 512, 7, 7]]      [1, 512, 7, 7]         2,048     
    Conv2D-314        [[1, 512, 7, 7]]     [1, 2048, 7, 7]       1,048,576   
  BatchNorm2D-314    [[1, 2048, 7, 7]]     [1, 2048, 7, 7]         8,192     
BottleneckBlock-32   [[1, 2048, 7, 7]]     [1, 2048, 7, 7]           0       
AdaptiveAvgPool2D-6  [[1, 2048, 7, 7]]     [1, 2048, 1, 1]           0       
     Linear-6           [[1, 2048]]             [1, 7]            14,343     
===============================================================================
Total params: 23,575,495
Trainable params: 23,469,255
Non-trainable params: 106,240
-------------------------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 261.48
Params size (MB): 89.93
Estimated Total Size (MB): 351.98
-------------------------------------------------------------------------------

C:\Users\Administrator\anaconda3\lib\site-packages\paddle\nn\layer\norm.py:635: UserWarning: When training, we now always track global mean and variance.
  warnings.warn(
2021-03-06 19:41:01,512 - INFO: Epoch:1/2, batch:1, train_loss:[2.08230], acc_top1:[0.18750], acc_top5:[0.79688](1.27s)
2021-03-06 19:41:02,223 - INFO: Epoch:1/2, batch:2, train_loss:[1.78195], acc_top1:[0.40625], acc_top5:[0.84375](0.71s)
2021-03-06 19:41:03,030 - INFO: Epoch:1/2, batch:3, train_loss:[1.75363], acc_top1:[0.25000], acc_top5:[0.84375](0.81s)
2021-03-06 19:41:03,836 - INFO: Epoch:1/2, batch:4, train_loss:[1.86290], acc_top1:[0.28125], acc_top5:[0.84375](0.81s)
2021-03-06 19:41:04,613 - INFO: Epoch:1/2, batch:5, train_loss:[1.40329], acc_top1:[0.48438], acc_top5:[0.93750](0.78s)
2021-03-06 19:41:05,378 - INFO: Epoch:1/2, batch:6, train_loss:[1.26327], acc_top1:[0.56250], acc_top5:[0.90625](0.76s)
2021-03-06 19:41:05,765 - INFO: [validation] Epoch:1/2, val_loss:[0.02363], val_top1:[0.53731], val_top5:[0.94030]
C:\Users\Administrator\anaconda3\lib\site-packages\paddle\fluid\layers\math_op_patch.py:293: UserWarning: C:\Users\Administrator\anaconda3\lib\site-packages\paddle\vision\models\resnet.py:145
The behavior of expression A + B has been unified with elementwise_add(X, Y, axis=-1) from Paddle 2.0. If your code works well in the older versions but crashes in this version, try to use elementwise_add(X, Y, axis=0) instead of A + B. This transitional warning will be dropped in the future.
  warnings.warn(
2021-03-06 19:41:09,185 - INFO: 已保存当前测试模型(epoch=1)为最优模型:Butterfly_Mobilenetv2_without_pretrained_final
2021-03-06 19:41:09,186 - INFO: 最优top1测试精度:0.53731 (epoch=1)
2021-03-06 19:41:10,063 - INFO: Epoch:2/2, batch:7, train_loss:[1.08590], acc_top1:[0.68750], acc_top5:[0.98438](4.69s)
2021-03-06 19:41:10,886 - INFO: Epoch:2/2, batch:8, train_loss:[0.86281], acc_top1:[0.73438], acc_top5:[0.96875](0.82s)
2021-03-06 19:41:11,659 - INFO: Epoch:2/2, batch:9, train_loss:[0.94885], acc_top1:[0.68750], acc_top5:[0.95312](0.77s)
2021-03-06 19:41:12,452 - INFO: Epoch:2/2, batch:10, train_loss:[0.55992], acc_top1:[0.85938], acc_top5:[1.00000](0.79s)
2021-03-06 19:41:13,242 - INFO: Epoch:2/2, batch:11, train_loss:[0.47196], acc_top1:[0.85938], acc_top5:[0.98438](0.79s)
2021-03-06 19:41:14,046 - INFO: Epoch:2/2, batch:12, train_loss:[0.55990], acc_top1:[0.79688], acc_top5:[0.95312](0.80s)
2021-03-06 19:41:14,417 - INFO: [validation] Epoch:2/2, val_loss:[0.00819], val_top1:[0.79104], val_top5:[0.98507]
2021-03-06 19:41:17,910 - INFO: 已保存当前测试模型(epoch=2)为最优模型:Butterfly_Mobilenetv2_without_pretrained_final
2021-03-06 19:41:17,911 - INFO: 最优top1测试精度:0.79104 (epoch=2)
2021-03-06 19:41:17,912 - INFO: 训练完成，最终性能accuracy=0.79104(epoch=2), 总耗时17.67s, 已将其保存为：Butterfly_Mobilenetv2_without_pretrained_final
2021-03-06 19:41:17,912 - INFO: 训练完毕，结果路径D:\Workspace\ExpResults\Butterfly_Mobilenetv2_without_pretrained_202103061920.
2021-03-06 19:41:17,913 - INFO: Done.


                        import os
import time
import json
from config import *
from utils.optimizer import learning_rate_setting, optimizer_setting
from utils.logger import logger
from utils.reader import dataset_train, dataset_trainval, dataset_val
import paddle
from paddle.static import InputSpec

def train(model):
    model.fit(train_data  = dataset_train,
              eval_data   = dataset_val,
              epochs      = train_parameters['total_epoch'],
              batch_size  = train_parameters['batch_size'],
              log_freq    = train_parameters['log_interval'],
              eval_freq   = train_parameters['eval_interval'],
            #   save_dir    = checkpoint_path,      # checkpoint model用于恢复训练
              save_freq   = train_parameters['total_epoch'],
              shuffle     = True, 
              verbose     = 1)                   # 1 折叠每个epoch的log，2打开每个epoch的log
    
    # finetune model 用于调优和恢复训练
    model.save(os.path.join(checkpoint_path, model_name + '_final'))          
    # inference model 用于部署和预测                                                  
    model.save(os.path.join(final_models_path, model_name + '_final'), training=False)
    
    
if __name__ == '__main__':
    # 将此次训练的超参数进行保存
    data = json.dumps(train_parameters, indent=4, ensure_ascii=False, sort_keys=False, separators=(',', ':'))   # 格式化字典格式的参数列表
    logger.info(data)
    # 启动训练过程
    logger.info('训练参数保存完毕，使用{}模型, 训练{}数据, 训练集{}, 启动训练...'.format(train_parameters['architecture'],train_parameters['dataset_name'],train_parameters['training_data']))
    logger.info('当前模型目录为：{}'.format(model_name + '_' + train_parameters['starting_time']))

    # 设置输入样本的维度
    input_spec = InputSpec(shape=[None] + train_parameters['input_size'], dtype='float32', name='image')
    label_spec = InputSpec(shape=[None, 1], dtype='int64', name='label')
  
    # 载入官方标准模型，若不存在则会自动进行下载，pretrained=True|False控制是否使用Imagenet预训练参数
    network = paddle.vision.models.resnet50(num_classes=train_parameters['class_dim'], pretrained=train_parameters['pretrained'])  
    model = paddle.Model(network, input_spec, label_spec)
    logger.info('模型参数信息：')
    # logger.info(model.summary()) # 是否显示神经网络的具体信息
    
    if train_parameters['checkpoint_train'] == True:
        model.load(checkpoint_load_model)
        logger.info('载入{}中断模型和参数完毕，开始从checkpoint恢复训练'.format(train_parameters['architecture']))
        logger.info('checkpoint模型:{}'.format(checkpoint_load_model))
    else:
        if train_parameters['pretrained'] == False:
            logger.info('载入{}模型完毕，从初始状态开始训练'.format(train_parameters['architecture']))
        elif train_parameters['pretrained_model'] == 'API':
            logger.info('载入Imagenet-{}预训练模型完毕，开始微调训练(fine-tune)'.format(train_parameters['architecture']))
        else:
            model.load(project_pretrained_model)
            logger.info('载入自定义预训练{}模型完毕，开始微调训练(fine-tune)'.format(train_parameters['architecture']))
            logger.info('预训练模型:{}'.format(project_pretrained_model))
          
    # 设置学习率、优化器、损失函数和评价指标
    lr = learning_rate_setting()
    optimizer = optimizer_setting(model, lr)    
    model.prepare(optimizer,
                  paddle.nn.CrossEntropyLoss(),
                  paddle.metric.Accuracy(topk=(1,5)))   

    # 启动训练过程
    start = time.perf_counter()
    train(model)
    logger.info('训练完毕，结果路径{}.'.format(result_root_path))

    print('Done.')

C:\Users\Administrator\anaconda3\lib\site-packages\ipykernel\ipkernel.py:287: DeprecationWarning: `should_run_async` will not call `transform_cell` automatically in the future. Please pass the result to `transformed_cell` argument and any exception that happen during thetransform in `preprocessing_exc_tuple` in IPython 7.17 and above.
  and should_run_async(code)
2021-03-06 19:16:07,193 - INFO: {
    "dataset_name":"Butterfly",
    "architecture":"Mobilenetv2_without",
    "training_data":"train",
    "starting_time":"202103061915",
    "input_size":[
        3,
        224,
        224
    ],
    "mean_value":[
        0.485,
        0.456,
        0.406
    ],
    "std_value":[
        0.229,
        0.224,
        0.225
    ],
    "num_trainval":490,
    "num_train":423,
    "num_val":67,
    "num_test":129,
    "class_dim":7,
    "label_dict":{
        "0":"admiral",
        "1":"black_swallowtail",
        "2":"machaon",
        "3":"monarch_closed",
        "4":"monarch_open",
        "5":"peacock",
        "6":"zebra"
    },
    "total_epoch":2,
    "batch_size":64,
    "log_interval":1,
    "eval_interval":1,
    "checkpointed":false,
    "checkpoint_train":false,
    "checkpoint_model":"Butterfly_Mobilenetv2_pretrained_final",
    "checkpoint_time":"202102182058",
    "pretrained":true,
    "pretrained_model":"API",
    "dataset_root_path":"D:\\Workspace\\ExpDatasets",
    "result_root_path":"D:\\Workspace\\ExpResults",
    "project_result_path":"D:\\Workspace\\MyProjects\\ButterflyClassification\\results",
    "useGPU":true,
    "learning_strategy":{
        "optimizer_strategy":"Momentum",
        "learning_rate_strategy":"CosineAnnealingDecay",
        "learning_rate":0.01,
        "momentum":0.9,
        "Piecewise_boundaries":[
            60,
            80,
            90
        ],
        "Piecewise_values":[
            0.01,
            0.001,
            0.0001,
            1e-05
        ],
        "Exponential_gamma":0.9,
        "Polynomial_decay_steps":10,
        "verbose":false
    },
    "augmentation_strategy":{
        "withAugmentation":true,
        "augmentation_prob":0.5,
        "rotate_angle":15,
        "Hflip_prob":0.5,
        "brightness":0.4,
        "contrast":0.4,
        "saturation":0.4,
        "hue":0.4
    }
}
2021-03-06 19:16:07,194 - INFO: 训练参数保存完毕，使用Mobilenetv2_without模型, 训练Butterfly数据, 训练集train, 启动训练...
2021-03-06 19:16:07,195 - INFO: 当前模型目录为：Butterfly_Mobilenetv2_without_202103061915
2021-03-06 19:16:07,272 - INFO: unique_endpoints {''}
2021-03-06 19:16:07,273 - INFO: File C:\Users\Administrator/.cache/paddle/hapi/weights\resnet50.pdparams md5 checking...
2021-03-06 19:16:07,566 - INFO: Found C:\Users\Administrator/.cache/paddle/hapi/weights\resnet50.pdparams
2021-03-06 19:16:08,269 - INFO: 模型参数信息：
2021-03-06 19:16:08,269 - INFO: 载入Imagenet-Mobilenetv2_without预训练模型完毕，开始微调训练(fine-tune)

The loss value printed in the log is the current step, and the metric is the average value of previous step.
Epoch 1/2
step 7/7 [==============================] - loss: 1.0073 - acc_top1: 0.5485 - acc_top5: 0.9007 - 632ms/step          
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 2/2 [==============================] - loss: 0.8268 - acc_top1: 0.8507 - acc_top5: 0.9254 - 198ms/step          
Eval samples: 67
Epoch 2/2
step 7/7 [==============================] - loss: 0.1625 - acc_top1: 0.9078 - acc_top5: 0.9858 - 564ms/step          
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 2/2 [==============================] - loss: 5.4584e-04 - acc_top1: 0.9104 - acc_top5: 1.0000 - 202ms/step      
Eval samples: 67

2021-03-06 19:16:21,244 - INFO: 训练完毕，结果路径D:\Workspace\ExpResults\Butterfly_Mobilenetv2_without_202103061915.

Done.


                        from utils.reader import dataset_val
result = model.evaluate(dataset_val, batch_size=args['batch_size'],verbose=0)
print('验证集top1准确率:{:.5f}, top5准确率:{:.5f}'.format(result['acc_top1'], result['acc_top5']))

验证集top1准确率:0.91045, top5准确率:1.00000


                        from utils.reader import dataset_test
result = model.evaluate(dataset_test, batch_size=args['batch_size'],verbose=0)
print('验证集top1准确率:{:.5f}, top5准确率:{:.5f}'.format(result['acc_top1'], result['acc_top5']))

验证集top1准确率:0.90698, top5准确率:1.00000


                        import numpy as np
import paddle.nn.functional as F

logits = model.predict(dataset_test, batch_size=args['batch_size'], stack_outputs=True)
pred = F.softmax(paddle.to_tensor(logits[0]))
pred_id = np.argmax(pred.numpy(), axis=1)
pred_id

Predict begin...
step 3/3 [==============================] - 253ms/step          
Predict samples: 129

array([0, 0, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
       2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
       3, 4, 4, 4, 3, 3, 4, 3, 4, 4, 4, 3, 3, 3, 3, 3, 3, 5, 5, 5, 5, 5,
       5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
       5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6],
      dtype=int64)


                        # 载入项目文件夹
import sys
sys.path.append(r"D:\\Workspace\\MyProjects\\ButterflyClassification")
#################################################################################


import numpy as np
import paddle
from config import train_parameters as args
from config import project_checkpoint_path
from utils.reader import val_reader, test_reader
import paddle.nn.functional as F
from paddle.static import InputSpec

__all__ = ['eval']

def eval(model, data_reader, verbose=0):
    if verbose == 1:
        print('开始评估...损失和精度均在上一个batch进行评估')

    accuracies_top1 = []
    accuracies_top5 = []
    losses = []
    n_total = 0
    
    for batch_id, (image, label) in enumerate(data_reader):
        n_batch = len(label)
        n_total = n_total + n_batch

        label = paddle.unsqueeze(label, axis=1)

        loss, acc = model.eval_batch([image], [label])
        losses.append(loss[0])                    
        accuracies_top1.append(acc[0][0]*n_batch)
        accuracies_top5.append(acc[0][1]*n_batch) 

        if verbose == 1:
            print('Batch:{}/{}, acc_top1:[{:.5f}], acc_top5:[{:.5f}]'.format(batch_id+1, len(data_reader), acc[0][0], acc[0][1])) 
                
    avg_loss = np.sum(losses)/n_total                 # loss 记录的是当前batch的累积值
    avg_acc_top1 = np.sum(accuracies_top1)/n_total    # metric 是当前batch的平均值
    avg_acc_top5 = np.sum(accuracies_top5)/n_total

    return avg_loss, avg_acc_top1, avg_acc_top5

##############################################################     
if __name__ == '__main__':
    # 设置输入样本的维度    
    input_spec = InputSpec(shape=[None] + args['input_size'], dtype='float32', name='image')
    label_spec = InputSpec(shape=[None, 1], dtype='int64', name='label')
    
    # 载入模型
    network = paddle.vision.models.mobilenet_v2(num_classes=args['class_dim'])  # 载入模型结构
    model = paddle.Model(network, input_spec, label_spec)                   # 模型实例化
    model.load(project_checkpoint_path)                                     # 载入调优模型的参数
    model.prepare(loss=paddle.nn.CrossEntropyLoss(),                        # 设置loss
                  metrics=paddle.metric.Accuracy(topk=(1,5)))               # 设置评价指标
    
    # 执行评估函数，并输出验证集样本的损失和精度
    avg_loss, avg_acc_top1, avg_acc_top5 = eval(model, val_reader(), verbose=0)
    print('[验证集] 损失: {:.5f}, top1精度:{:.5f}, top5精度为:{:.5f}'.format(avg_loss, avg_acc_top1, avg_acc_top5))    
    avg_loss, avg_acc_top1, avg_acc_top5 = eval(model, test_reader())
    print('[测试集] 损失: {:.5f}, top1精度:{:.5f}, top5精度为:{:.5f}'.format(avg_loss, avg_acc_top1, avg_acc_top5))

[验证集] 损失: 0.00829, top1精度:0.88060, top5精度为:0.98507
[测试集] 损失: 0.03266, top1精度:0.84496, top5精度为:0.99225


                        # 使用微调模型对验证集和测试集进行评估

import paddle
from config import train_parameters as args
from config import project_checkpoint_path
from paddle.static import InputSpec
from utils.reader import *

def eval(dataset):
    result = model.evaluate(dataset, batch_size=args['batch_size'], verbose=0)

    return result['loss'][0], result['acc_top1'], result['acc_top5']

############################################################## 
if __name__ == '__main__':
    # 设置输入样本的维度    
    input_spec=InputSpec(shape=[None] + args['input_size'], dtype='float32', name='image')
    label_spec=InputSpec(shape=[None, 1], dtype='int64', name='label')
    
    # 载入模型
    network = paddle.vision.models.mobilenet_v2(num_classes=args['class_dim'])  # 载入模型结构
    model = paddle.Model(network, input_spec, label_spec)                   # 模型实例化
    model.load(project_checkpoint_path)                                     # 载入调优模型的参数
    model.prepare(loss=paddle.nn.CrossEntropyLoss(),                        # 设置loss
                  metrics=paddle.metric.Accuracy(topk=(1,5)))               # 设置评价指标
    
    # 执行评估函数，并输出验证集样本的损失和精度
    avg_loss, avg_acc_top1, avg_acc_top5 = eval(dataset_val)
    print('[验证集] 损失: {:.5f}, top1精度:{:.5f}, top5精度为:{:.5f}'.format(avg_loss, avg_acc_top1, avg_acc_top5))    
    avg_loss, avg_acc_top1, avg_acc_top5 = eval(dataset_test)
    print('[测试集] 损失: {:.5f}, top1精度:{:.5f}, top5精度为:{:.5f}'.format(avg_loss, avg_acc_top1, avg_acc_top5))

C:\Users\Administrator\anaconda3\lib\site-packages\paddle\fluid\layers\utils.py:77: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated since Python 3.3, and in 3.9 it will stop working
  return (isinstance(seq, collections.Sequence) and

[验证集] 损失: 0.00317, top1精度:0.88060, top5精度为:0.98507
[测试集] 损失: 3.00281, top1精度:0.84496, top5精度为:0.99225


                        import numpy as np
import paddle
from config import train_parameters as args
from config import project_final_model_path
from utils.reader import val_reader, test_reader
import paddle.nn.functional as F

def eval(model, data_reader):
    accuracies_top1 = []
    accuracies_top5 = []
    losses = []
    n_total = 0
    
    for batch_id, (image, label) in enumerate(data_reader):
        n_batch = len(label)
        n_total = n_total + n_batch

        label = paddle.unsqueeze(label, axis=1)       # 将label扩展为二阶张量

        logits = model(image)
        pred = F.softmax(logits)
        acc_top1 = paddle.metric.accuracy(pred, label, k=1)
        acc_top5 = paddle.metric.accuracy(pred, label, k=5)
        loss = F.cross_entropy(logits, label)
        avg_loss = paddle.mean(loss)  # 获取loss值

        losses.append(avg_loss.numpy())
        accuracies_top1.append(acc_top1.numpy()*n_batch) 
        accuracies_top5.append(acc_top5.numpy()*n_batch) 
    
    avg_loss = np.sum(losses)/n_total
    avg_acc_top1 = np.sum(accuracies_top1)/n_total
    avg_acc_top5 = np.sum(accuracies_top5)/n_total
    
    return avg_loss, avg_acc_top1, avg_acc_top5
    
##############################################################    
if __name__ == '__main__':
    # 载入模型
    model = paddle.jit.load(project_final_model_path)
    
    # 执行评估函数，并输出验证集样本的损失和精度
    avg_loss, avg_acc_top1, avg_acc_top5 = eval(model, val_reader())
    print('[验证集] 损失: {:.5f}, top1精度:{:.5f}, top5精度为:{:.5f}'.format(avg_loss, avg_acc_top1, avg_acc_top5))    
    avg_loss, avg_acc_top1, avg_acc_top5 = eval(model, test_reader())
    print('[测试集] 损失: {:.5f}, top1精度:{:.5f}, top5精度为:{:.5f}'.format(avg_loss, avg_acc_top1, avg_acc_top5))

[验证集] 损失: 0.00829, top1精度:0.88060, top5精度为:0.98507
[测试集] 损失: 0.03266, top1精度:0.84496, top5精度为:0.99225

蝴蝶分类(Butterfly Classification)¶

一、使用ResNet50模型进行训练¶

1.1 标准训练模式¶

trainval.py¶

1.2 使用高层API接口进行训练¶

trainvalAPI.py¶

1.3 基于训练好的模型(高层API接口)直接对验证集和测试集进行评估¶

1.3.1 验证集准确率¶

1.3.2 测试集准确率¶

1.3.3 输出预测结果¶

二、验证集和测试集评估¶

2.1 （推荐）使用调优模型进行评估（可以在训练中进行调用）¶

eval.py¶

2.2 使用高层API接口在调优模型上进行评估¶

evalAPI.py¶

2.3 使用部署模型进行评估¶

evalDeployed.py¶