Paddle入门实战系列：基于PaddleOCR的车牌识别

Posted 2022-12-07 GoAI

tags:

篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了Paddle入门实战系列：基于PaddleOCR的车牌识别相关的知识，希望对你有一定的参考价值。

✨写在前面：强烈推荐给大家一个优秀的人工智能学习网站，内容包括人工智能基础、机器学习、深度学习神经网络等，详细介绍各部分概念及实战教程，通俗易懂，非常适合人工智能领域初学者及研究者学习。➡️点击跳转到网站。

项目背景

车牌识别技术是智能交通的重要环节，目前已广泛应用于例如停车场、收费站等等交通设施中，提供高效便捷的车辆认证的服务，其中较为典型的应用场景为卡口系统。车牌识别即识别车牌上的文字信息，属于光学字符识别（OCR）的一项子任务。

项目流程

本项目基于Paddleocr完成深度学习车牌识别，项目可分为车牌检测与车牌识别两部分，主要流程为数据预处理、模型训练、导出推理模型、测试。

检测车牌所在图片位置
识别车牌图片具体内容

项目链接：基于Paddle的智慧交通预测系统 - 飞桨AI Studio

PaddleOCR介绍

主要由DB文本检测、检测框矫正和CRNN文本识别三部分组成。该系统从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身(如绿框所示)，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。

数据集介绍

本项目数据集采用CCPD2019和CCPD02020车牌数据集，以CCPD02020演示车牌识别流程，其训练、验证、测试数据集已划分完毕，测试集包含5006张图片，大小共865.66M。本项目演示的CCPD20数据集的采集方式与CCPD19类似，其中均为新能源车辆的车牌,其命名为ccpd_green路径。

CCPD2020/ccpd_green/

--train

--val

--test

文件名即图片标注，具体含义可查看源网址，以下为数据集解读参考。

“025-95_113-154&383_386&473-386&473_177&454_154&383_363&402-0_0_22_27_27_33_16-37-15.jpg”.

数据集字段解释参考：

CCPD数据集没有专门的标注文件，每张图像的文件名就是对应的数据标注(label)

例如:025-95_113-154&383_386&473-386&473_177&454_154&383_363&402-0_0_22_27_27_33_16-37-15.jpg 由分隔符'-'分为几个部分:

025为区域
95_113 对应两个角度, 水平95°, 竖直113°
154&383_386&473对应边界框坐标:左上(154, 383), 右下(386, 473)
386&473_177&454_154&383_363&402对应四个角点坐标
0_0_22_27_27_33_16为车牌号码映射关系如下: 第一个为省份0 对应省份字典皖, 后面的为字母和文字, 查看ads字典.如0为A, 22为Y....

安装PaddleOCR环境

#下载PaddleOCR
%cd ~/
!git clone -b release/2.1 https://github.com/PaddlePaddle/PaddleOCR.git

#安装环境
%cd PaddleOCR
!pip install -r requirments.txt
!pip install --upgrade scipy
# !pip install -r requirements.txt -i https://mirror.baidu.com/pypi/simple

数据预处理

字典内容：

车牌省份： provinces = [“皖”, “沪”, “津”, “渝”, “冀”, “晋”, “蒙”, “辽”, “吉”, “黑”, “苏”, “浙”, “京”, “闽”, “赣”, “鲁”, “豫”, “鄂”, “湘”, “粤”, “桂”, “琼”, “川”, “贵”, “云”, “藏”, “陕”, “甘”, “青”, “宁”, “新”, “警”, “学”, “O”]

alphabets = [‘A’, ‘B’, ‘C’, ‘D’, ‘E’, ‘F’, ‘G’, ‘H’, ‘J’, ‘K’, ‘L’, ‘M’, ‘N’, ‘P’, ‘Q’, ‘R’, ‘S’, ‘T’, ‘U’, ‘V’, ‘W’, ‘X’, ‘Y’, ‘Z’, ‘O’]

ads = [‘A’, ‘B’, ‘C’, ‘D’, ‘E’, ‘F’, ‘G’, ‘H’, ‘J’, ‘K’, ‘L’, ‘M’, ‘N’, ‘P’, ‘Q’, ‘R’, ‘S’, ‘T’, ‘U’, ‘V’, ‘W’, ‘X’, ‘Y’, ‘Z’, ‘0’, ‘1’, ‘2’, ‘3’, ‘4’, ‘5’, ‘6’, ‘7’, ‘8’, ‘9’, ‘O’]

!unzip -q data/data168819/CCPD2020.zip -d work/CCPD2020

检测数据：转换成官方提供的icdar格式

格式如下：

transcription为具体内容，points存放检测车牌坐标

03706896551724138-90_263-185&516_545&610-543&607_194&610_185&518_545&516-0_0_3_29_27_33_24_33-59-41.jpg ["transcription": "皖AD53909", "points": [[185, 518], [545, 516], [543, 607], [194, 610]]] /034782088122605366-92_244-167&522_517&612-517&612_195&590_167&522_497&539-0_0_3_24_30_30_33_25-102-110.jpg ["transcription": "皖AD06691", "points": [[167, 522], [497, 539], [517, 612], [195, 590]]]

In [ ]

#转换检测数据，打开注释执行三次生成训练所需txt文件，分别为train、val、test。
%cd ~
import os, cv2
words_list = [
    "A", "B", "C", "D", "E",
    "F", "G", "H", "J", "K", 
    "L", "M", "N", "P", "Q", 
    "R", "S", "T", "U", "V", 
    "W", "X", "Y", "Z", "0", 
    "1", "2", "3", "4", "5", 
    "6", "7", "8", "9" ]

con_list = [
    "皖", "沪", "津", "渝", "冀",
    "晋", "蒙", "辽", "吉", "黑",
    "苏", "浙", "京", "闽", "赣",
    "鲁", "豫", "鄂", "湘", "粤",
    "桂", "琼", "川", "贵", "云",
    "西", "陕", "甘", "青", "宁",
    "新"]

count = 0
# data = open('work/train_data_det.txt', 'w', encoding='UTF-8')
# data = open('work/val_data_det.txt', 'w', encoding='UTF-8')
data = open('work/test_data_det.txt', 'w', encoding='UTF-8')

# for item in os.listdir('work/CCPD2020/CCPD2020/ccpd_green/train'):
# for item in os.listdir('work/CCPD2020/CCPD2020/ccpd_green/val'):
for item in os.listdir('work/CCPD2020/CCPD2020/ccpd_green/test'):

    # path = 'work/CCPD2020/CCPD2020/ccpd_green/train/'+item
    # path = 'work/CCPD2020/CCPD2020/ccpd_green/val/'+item
    path = 'work/CCPD2020/CCPD2020/ccpd_green/test/'+item

    _, _, bbox, points, label, _, _ = item.split('-')
    points = points.split('_')
    points = [_.split('&') for _ in points]
    tmp = points[-2:]+points[:2]
    points = []
    for point in tmp:
        points.append([int(_) for _ in point])
    label = label.split('_')
    con = con_list[int(label[0])]
    words = [words_list[int(_)] for _ in label[1:]]
    label = con+''.join(words)
    line = path+'\\t'+'["transcription": "%s", "points": %s]' % (label, str(points))
    line = line[:]+'\\n'
    data.write(line)

total = []
# with open('work/train_data_det.txt', 'r', encoding='UTF-8') as f:
#     for line in f:
#         total.append(line)

# with open('work/val_data_det.txt', 'r', encoding='UTF-8') as f:
#     for line in f:
#         total.append(line)

with open('work/test_data_det.txt', 'r', encoding='UTF-8') as f:
    for line in f:
        total.append(line)


# with open('work/train_det.txt', 'w', encoding='UTF-8') as f:
#     for line in total[:-500]:
#         f.write(line)

# with open('work/dev_det.txt', 'w', encoding='UTF-8') as f:
#     for line in total[-500:]:
#         f.write(line)

/home/aistudio

In [ ]

#识别数据：转换成PaddleOCR使用的格式（图片名+内容），打开注释执行三次生成训练所需txt文件，分别为train、val、test。
%cd ~
import os, cv2
words_list = [
    "A", "B", "C", "D", "E",
    "F", "G", "H", "J", "K", 
    "L", "M", "N", "P", "Q", 
    "R", "S", "T", "U", "V", 
    "W", "X", "Y", "Z", "0", 
    "1", "2", "3", "4", "5", 
    "6", "7", "8", "9" ]

con_list = [
    "皖", "沪", "津", "渝", "冀",
    "晋", "蒙", "辽", "吉", "黑",
    "苏", "浙", "京", "闽", "赣",
    "鲁", "豫", "鄂", "湘", "粤",
    "桂", "琼", "川", "贵", "云",
    "西", "陕", "甘", "青", "宁",
    "新"]

# if not os.path.exists('work/img'):   #所有数据集都放入一个文件夹
#     os.mkdir('work/img')

#训练、验证、测试集分开三个文件夹对应解开注释依次执行三次
# if not os.path.exists('work/train_rec_img'):
#     os.mkdir('work/train_rec_img')


if not os.path.exists('work/val_rec_img'):
    os.mkdir('work/val_rec_img')

# if not os.path.exists('work/test_rec_img'):
#     os.mkdir('work/test_rec_img')

count = 0

# data = open('work/train_data_rec.txt', 'w', encoding='UTF-8')
data = open('work/val_data_rec.txt', 'w', encoding='UTF-8')
# data = open('work/test_data_rec.txt', 'w', encoding='UTF-8')

# for item in os.listdir('work/CCPD2020/CCPD2020/ccpd_green/train'):
for item in os.listdir('work/CCPD2020/CCPD2020/ccpd_green/val'):
# for item in os.listdir('work/CCPD2020/CCPD2020/ccpd_green/test'):

    # path = 'work/CCPD2020/CCPD2020/ccpd_green/train/'+item
    path = 'work/CCPD2020/CCPD2020/ccpd_green/val/'+item
    # path = 'work/CCPD2020/CCPD2020/ccpd_green/test/'+item
    
    #原来的 path = 'work/CCPD2020/ccpd_base/'+item

    _, _, bbox, _, label, _, _ = item.split('-')
    bbox = bbox.split('_')
    x1, y1 = bbox[0].split('&')
    x2, y2 = bbox[1].split('&')
    label = label.split('_')
    con = con_list[int(label[0])]
    words = [words_list[int(_)] for _ in label[1:]]
    label = con+''.join(words)
    bbox = [int(_) for _ in [x1, y1, x2, y2]]
    img = cv2.imread(path)
    crop = img[bbox[1]:bbox[3], bbox[0]:bbox[2], :]

    # cv2.imwrite('work/train_rec_img/%06d.jpg' % count, crop)
    # data.write('work/train_rec_img/%06d.jpg\\t%s\\n' % (count, label))

    cv2.imwrite('work/val_rec_img/%06d.jpg' % count, crop)
    data.write('work/val_rec_img/%06d.jpg\\t%s\\n' % (count, label))
    
    # cv2.imwrite('work/test_rec_img/%06d.jpg' % count, crop)
    # data.write('work/test_rec_img/%06d.jpg\\t%s\\n' % (count, label))
    count += 1
data.close()

with open('work/word_dict.txt', 'w', encoding='UTF-8') as f:
    for line in words_list+con_list:
        f.write(line+'\\n')

# total = []
# with open('work/train_data_rec.txt', 'r', encoding='UTF-8') as f:
#     for line in f:
#         total.append(line)

with open('work/val_data_rec.txt', 'r', encoding='UTF-8') as f:
    for line in f:
        total.append(line)

# with open('work/test_data_rec.txt', 'r', encoding='UTF-8') as f:
#     for line in f:
#         total.append(line)

/home/aistudio

最终文件生成效果：

注：”由于work文件较大，无法生成版本，读者可执行上述代码生成下述对应文件,也可以自行修改代码，放到其他路径。
下图为work目录下，其中CCPD为本项目CCPD2020数据（解压到work目录下），另三个rec_img文件夹为用于识别流程的图片，img为三个汇总，其余为用于检测det与识别rec生成的txt文本。

模型介绍

PaddleOCR提供的检测与识别模型如下：

模型选择

Paddocr提供以下文件检测及识别模型，考虑车牌识别中字符数量较少，而且长度也固定，且为标准的印刷字体，所以无需使用过于复杂的模型。因此，参考其他开源资料，本项目选择经典的DBNet+RARE，两个模型均使用MobileNetV3作为其主干网络(Backbone)。

DBNet与RARE算法介绍可参考我的博客：OCR文字识别经典论文详解

DBNet

RARE

下载对应预训练模型

In [ ]

%cd ~/work/PaddleOCR
# 下载预训练模型
!wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ch_models/ch_det_mv3_db.tar
!wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/rec_mv3_tps_bilstm_attn.tar
# 解压模型参数
%cd pretrain_models
!tar -xf ch_det_mv3_db.tar && rm -rf ch_det_mv3_db.tar
!tar -xf rec_mv3_tps_bilstm_attn.tar && rm -rf rec_mv3_tps_bilstm_attn.tar

模型训练

1.检测模型训练

检测模版文件 configs/det/det_mv3_db.yml

Global:
  algorithm: DB
  use_gpu: true
  epoch_num: 1200
  log_smooth_window: 20
  print_batch_step: 20
  save_model_dir: ./myoutput/det_db/
  save_epoch_step: 10
  # evaluation is run every 5000 iterations after the 4000th iteration
  eval_batch_step: [100, 500]
  train_batch_size_per_card: 4
  test_batch_size_per_card: 4
  image_shape: [3, 640, 640]
  reader_yml: ./configs/det/det_db_icdar15_reader.yml
  pretrain_weights: ./pretrain_models/det_mv3_db/best_accuracy
  checkpoints:
  save_res_path: ./myoutput/det_db/predicts_db.txt
  save_inference_dir:
  
Architecture:
  function: ppocr.modeling.architectures.det_model,DetModel

Backbone:
  function: ppocr.modeling.backbones.det_mobilenet_v3,MobileNetV3
  scale: 0.5
  model_name: large

Head:
  function: ppocr.modeling.heads.det_db_head,DBHead
  model_name: large
  k: 50
  inner_channels: 96
  out_channels: 2

Loss:
  function: ppocr.modeling.losses.det_db_loss,DBLoss
  balance_loss: true
  main_loss_type: DiceLoss
  alpha: 5
  beta: 10
  ohem_ratio: 3

Optimizer:
  function: ppocr.optimizer,AdamDecay
  base_lr: 0.001
  beta1: 0.9
  beta2: 0.999

PostProcess:
  function: ppocr.postprocess.db_postprocess,DBPostProcess
  thresh: 0.3
  box_thresh: 0.7
  max_candidates: 1000
  unclip_ratio: 2.0

In [ ]

%cd ~/PaddleOCR
# 设置PYTHONPATH路径
%env PYTHONPATH=$PYTHONPATH:.
# GPU单卡训练
%env CUDA_VISIBLE_DEVICES=0
!python3 tools/train.py -c configs/det/det_mv3_db.yml

2.识别

主干网络为轻量级网络MobilenetV3,识别算法包括TPS校正+双向LSTM+Attention

Global:
  algorithm: RARE
  use_gpu: true
  epoch_num: 200
  log_smooth_window: 20
  print_batch_step: 20
  #save_model_dir: output/rec_RARE
  save_model_dir: ./myoutput/rec_RARE_atten_new
  save_epoch_step: 100
  eval_batch_step: 500
  train_batch_size_per_card: 256
  test_batch_size_per_card: 256
  image_shape: [3, 32, 320]
  max_text_length: 8
  character_type: ch
  character_dict_path: ../word_dict.txt
  loss_type: attention
  tps: true
  reader_yml: ./configs/rec/rec_chinese_reader.yml
  pretrain_weights: ./pretrain_models/rec_mv3_tps_bilstm_attn/best_accuracy
  # pretrain_weights:
  checkpoints: 
  save_inference_dir: ./inference/newimg_rec_rare
  infer_img:


Architecture:
  function: ppocr.modeling.architectures.rec_model,RecModel

TPS:
  function: ppocr.modeling.stns.tps,TPS
  num_fiducial: 20
  loc_lr: 0.1
  model_name: small
  
Backbone:
  function: ppocr.modeling.backbones.rec_mobilenet_v3,MobileNetV3
  scale: 0.5
  model_name: large
 
Head:
  function: ppocr.modeling.heads.rec_attention_head,AttentionPredict
  encoder_type: rnn
  SeqRNN:
    hidden_size: 96
  Attention:
    decoder_size: 96
    word_vector_dim: 96
  
Loss:
  function: ppocr.modeling.losses.rec_attention_loss,AttentionLoss
  
Optimizer:
  function: ppocr.optimizer,AdamDecay
  base_lr: 0.001
  beta1: 0.9
  beta2: 0.999

In [77]

%cd ~/PaddleOCR
# GPU单卡训练
%env CUDA_VISIBLE_DEVICES=0
!python3 tools/train.py -c configs/rec/rec_mv3_tps_bilstm_attn.yml

/home/aistudio/work/PaddleOCR
env: CUDA_VISIBLE_DEVICES=0
2022-09-15 16:06:31,610-INFO: 'Global': 'debug': False, 'algorithm': 'RARE', 'use_gpu': True, 'epoch_num': 200, 'log_smooth_window': 20, 'print_batch_step': 20, 'save_model_dir': './myoutput/rec_RARE_atten_new', 'save_epoch_step': 100, 'eval_batch_step': 500, 'train_batch_size_per_card': 256, 'test_batch_size_per_card': 256, 'image_shape': [3, 32, 320], 'max_text_length': 8, 'character_type': 'ch', 'character_dict_path': '../word_dict.txt', 'loss_type': 'attention', 'tps': True, 'reader_yml': './configs/rec/rec_chinese_reader.yml', 'pretrain_weights': './pretrain_models/rec_mv3_tps_bilstm_attn/best_accuracy', 'checkpoints': None, 'save_inference_dir': './inference/newimg_rec_rare', 'infer_img': None, 'Architecture': 'function': 'ppocr.modeling.architectures.rec_model,RecModel', 'TPS': 'function': 'ppocr.modeling.stns.tps,TPS', 'num_fiducial': 20, 'loc_lr': 0.1, 'model_name': 'small', 'Backbone': 'function': 'ppocr.modeling.backbones.rec_mobilenet_v3,MobileNetV3', 'scale': 0.5, 'model_name': 'large', 'Head': 'function': 'ppocr.modeling.heads.rec_attention_head,AttentionPredict', 'encoder_type': 'rnn', 'SeqRNN': 'hidden_size': 96, 'Attention': 'decoder_size': 96, 'word_vector_dim': 96, 'Loss': 'function': 'ppocr.modeling.losses.rec_attention_loss,AttentionLoss', 'Optimizer': 'function': 'ppocr.optimizer,AdamDecay', 'base_lr': 0.001, 'beta1': 0.9, 'beta2': 0.999, 'TrainReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader', 'num_workers': 8, 'img_set_dir': '../..', 'label_file_path': '../train_data_rec.txt', 'EvalReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader', 'num_workers': 8, 'img_set_dir': '../..', 'label_file_path': '../val_data_rec.txt', 'TestReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader'
2022-09-15 16:06:32,400-INFO: If regularizer of a Parameter has been set by 'fluid.ParamAttr' or 'fluid.WeightNormParamAttr' already. The Regularization[L2Decay, regularization_coeff=0.000000] in Optimizer will not take effect, and it will only be applied to other Parameters!
2022-09-15 16:06:34,241-INFO: places would be ommited when DataLoader is not iterable
W0915 16:06:34.303215 18580 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 11.2, Runtime API Version: 9.0
W0915 16:06:34.307891 18580 device_context.cc:260] device: 0, cuDNN Version: 7.6.
2022-09-15 16:06:36,359-INFO: Loading parameters from ./pretrain_models/rec_mv3_tps_bilstm_attn/best_accuracy...
2022-09-15 16:06:36,425-WARNING: variable embedding_0.w_0 not used
2022-09-15 16:06:36,425-WARNING: variable rnn_out_fc.w_0 not used
2022-09-15 16:06:36,425-WARNING: variable rnn_out_fc.b_0 not used
2022-09-15 16:06:36,470-INFO: Finish initing model from ./pretrain_models/rec_mv3_tps_bilstm_attn/best_accuracy
2022-09-15 16:06:47,710-INFO: epoch: 1, iter: 20, lr: 0.001000, 'loss': 4937.542, 'acc': 0.0, time: 0.387
2022-09-15 16:06:56,909-INFO: epoch: 2, iter: 40, lr: 0.001000, 'loss': 3201.4744, 'acc': 0.0, time: 0.373
2022-09-15 16:07:06,105-INFO: epoch: 3, iter: 60, lr: 0.001000, 'loss': 2400.3447, 'acc': 0.007812, time: 0.374
2022-09-15 16:07:16,551-INFO: epoch: 5, iter: 80, lr: 0.001000, 'loss': 1625.6213, 'acc': 0.146484, time: 1.787
2022-09-15 16:07:25,677-INFO: epoch: 6, iter: 100, lr: 0.001000, 'loss': 740.44006, 'acc': 0.769531, time: 0.376
2022-09-15 16:07:34,829-INFO: epoch: 7, iter: 120, lr: 0.001000, 'loss': 370.51062, 'acc': 0.902344, time: 0.374
2022-09-15 16:07:43,827-INFO: epoch: 8, iter: 140, lr: 0.001000, 'loss': 281.9944, 'acc': 0.929688, time: 0.374
2022-09-15 16:07:54,266-INFO: epoch: 10, iter: 160, lr: 0.001000, 'loss': 213.27815, 'acc': 0.933594, time: 1.769
2022-09-15 16:08:03,481-INFO: epoch: 11, iter: 180, lr: 0.001000, 'loss': 149.04875, 'acc': 0.949219, time: 0.403
2022-09-15 16:08:12,440-INFO: epoch: 12, iter: 200, lr: 0.001000, 'loss': 119.06343, 'acc': 0.951172, time: 0.374
2022-09-15 16:08:21,519-INFO: epoch: 13, iter: 220, lr: 0.001000, 'loss': 100.26872, 'acc': 0.949219, time: 0.373
2022-09-15 16:08:32,031-INFO: epoch: 15, iter: 240, lr: 0.001000, 'loss': 84.79035, 'acc': 0.970703, time: 1.806
2022-09-15 16:08:41,146-INFO: epoch: 16, iter: 260, lr: 0.001000, 'loss': 79.10332, 'acc': 0.966797, time: 0.409
2022-09-15 16:08:50,249-INFO: epoch: 17, iter: 280, lr: 0.001000, 'loss': 97.50918, 'acc': 0.955078, time: 0.373
2022-09-15 16:08:59,211-INFO: epoch: 18, iter: 300, lr: 0.001000, 'loss': 81.51908, 'acc': 0.960938, time: 0.373
2022-09-15 16:09:09,670-INFO: epoch: 20, iter: 320, lr: 0.001000, 'loss': 153.25719, 'acc': 0.935547, time: 1.713
2022-09-15 16:09:18,849-INFO: epoch: 21, iter: 340, lr: 0.001000, 'loss': 122.89736, 'acc': 0.949219, time: 0.399
2022-09-15 16:09:28,010-INFO: epoch: 22, iter: 360, lr: 0.001000, 'loss': 78.92976, 'acc': 0.958984, time: 0.373
2022-09-15 16:09:37,014-INFO: epoch: 23, iter: 380, lr: 0.001000, 'loss': 53.346596, 'acc': 0.972656, time: 0.373
2022-09-15 16:09:47,468-INFO: epoch: 25, iter: 400, lr: 0.001000, 'loss': 52.11016, 'acc': 0.972656, time: 1.756
2022-09-15 16:09:56,459-INFO: epoch: 26, iter: 420, lr: 0.001000, 'loss': 43.66102, 'acc': 0.972656, time: 0.374
2022-09-15 16:10:05,548-INFO: epoch: 27, iter: 440, lr: 0.001000, 'loss': 38.42436, 'acc': 0.980469, time: 0.373
2022-09-15 16:10:14,536-INFO: epoch: 28, iter: 460, lr: 0.001000, 'loss': 45.512665, 'acc': 0.978516, time: 0.373
2022-09-15 16:10:25,020-INFO: epoch: 30, iter: 480, lr: 0.001000, 'loss': 40.709595, 'acc': 0.976562, time: 1.789
2022-09-15 16:10:34,337-INFO: epoch: 31, iter: 500, lr: 0.001000, 'loss': 39.14653, 'acc': 0.974609, time: 0.412
2022-09-15 16:10:36,277-INFO: Already save model in ./myoutput/rec_RARE_atten_new/best_accuracy
2022-09-15 16:10:36,277-INFO: Test iter: 500, acc:0.777223, best_acc:0.777223, best_epoch:31, best_batch_id:500, eval_sample_num:1001
2022-09-15 16:10:45,402-INFO: epoch: 32, iter: 520, lr: 0.001000, 'loss': 33.690575, 'acc': 0.984375, time: 0.377
2022-09-15 16:10:54,503-INFO: epoch: 33, iter: 540, lr: 0.001000, 'loss': 29.822884, 'acc': 0.982422, time: 0.374
2022-09-15 16:11:05,027-INFO: epoch: 35, iter: 560, lr: 0.001000, 'loss': 33.394993, 'acc': 0.980469, time: 1.868
2022-09-15 16:11:14,148-INFO: epoch: 36, iter: 580, lr: 0.001000, 'loss': 29.375713, 'acc': 0.980469, time: 0.419
2022-09-15 16:11:23,142-INFO: epoch: 37, iter: 600, lr: 0.001000, 'loss': 44.640976, 'acc': 0.976562, time: 0.374
2022-09-15 16:11:32,155-INFO: epoch: 38, iter: 620, lr: 0.001000, 'loss': 32.55059, 'acc': 0.988281, time: 0.373
2022-09-15 16:11:42,665-INFO: epoch: 40, iter: 640, lr: 0.001000, 'loss': 24.32478, 'acc': 0.988281, time: 1.872
2022-09-15 16:11:51,777-INFO: epoch: 41, iter: 660, lr: 0.001000, 'loss': 23.933027, 'acc': 0.986328, time: 0.375
2022-09-15 16:12:00,873-INFO: epoch: 42, iter: 680, lr: 0.001000, 'loss': 24.20562, 'acc': 0.990234, time: 0.376
2022-09-15 16:12:09,940-INFO: epoch: 43, iter: 700, lr: 0.001000, 'loss': 20.969837, 'acc': 0.988281, time: 0.374
2022-09-15 16:12:20,566-INFO: epoch: 45, iter: 720, lr: 0.001000, 'loss': 21.603209, 'acc': 0.988281, time: 1.798
2022-09-15 16:12:29,785-INFO: epoch: 46, iter: 740, lr: 0.001000, 'loss': 14.918937, 'acc': 0.992188, time: 0.380
2022-09-15 16:12:38,854-INFO: epoch: 47, iter: 760, lr: 0.001000, 'loss': 16.47085, 'acc': 0.992188, time: 0.373
2022-09-15 16:12:47,954-INFO: epoch: 48, iter: 780, lr: 0.001000, 'loss': 13.787853, 'acc': 0.992188, time: 0.373
2022-09-15 16:12:58,526-INFO: epoch: 50, iter: 800, lr: 0.001000, 'loss': 20.16197, 'acc': 0.990234, time: 1.886
2022-09-15 16:13:07,622-INFO: epoch: 51, iter: 820, lr: 0.001000, 'loss': 15.042562, 'acc': 0.992188, time: 0.413
2022-09-15 16:13:16,744-INFO: epoch: 52, iter: 840, lr: 0.001000, 'loss': 30.320072, 'acc': 0.980469, time: 0.374
2022-09-15 16:13:25,803-INFO: epoch: 53, iter: 860, lr: 0.001000, 'loss': 27.366596, 'acc': 0.984375, time: 0.373
2022-09-15 16:13:36,263-INFO: epoch: 55, iter: 880, lr: 0.001000, 'loss': 107.29559, 'acc': 0.964844, time: 1.804
2022-09-15 16:13:45,315-INFO: epoch: 56, iter: 900, lr: 0.001000, 'loss': 112.5357, 'acc': 0.9375, time: 0.377
2022-09-15 16:13:54,380-INFO: epoch: 57, iter: 920, lr: 0.001000, 'loss': 80.11721, 'acc': 0.951172, time: 0.375
2022-09-15 16:14:03,521-INFO: epoch: 58, iter: 940, lr: 0.001000, 'loss': 173.3816, 'acc': 0.896484, time: 0.375
2022-09-15 16:14:14,176-INFO: epoch: 60, iter: 960, lr: 0.001000, 'loss': 111.91419, 'acc': 0.943359, time: 1.820
2022-09-15 16:14:23,354-INFO: epoch: 61, iter: 980, lr: 0.001000, 'loss': 67.807884, 'acc': 0.964844, time: 0.418
2022-09-15 16:14:32,359-INFO: epoch: 62, iter: 1000, lr: 0.001000, 'loss': 49.46469, 'acc': 0.972656, time: 0.382
2022-09-15 16:14:34,236-INFO: Already save model in ./myoutput/rec_RARE_atten_new/best_accuracy
2022-09-15 16:14:34,237-INFO: Test iter: 1000, acc:0.782218, best_acc:0.782218, best_epoch:62, best_batch_id:1000, eval_sample_num:1001
2022-09-15 16:14:43,256-INFO: epoch: 63, iter: 1020, lr: 0.001000, 'loss': 43.5632, 'acc': 0.972656, time: 0.373
2022-09-15 16:14:53,769-INFO: epoch: 65, iter: 1040, lr: 0.001000, 'loss': 31.788738, 'acc': 0.984375, time: 1.797
2022-09-15 16:15:02,909-INFO: epoch: 66, iter: 1060, lr: 0.001000, 'loss': 24.396, 'acc': 0.984375, time: 0.411
2022-09-15 16:15:12,029-INFO: epoch: 67, iter: 1080, lr: 0.001000, 'loss': 21.756264, 'acc': 0.992188, time: 0.374
2022-09-15 16:15:21,012-INFO: epoch: 68, iter: 1100, lr: 0.001000, 'loss': 18.515491, 'acc': 0.988281, time: 0.379
2022-09-15 16:15:31,652-INFO: epoch: 70, iter: 1120, lr: 0.001000, 'loss': 28.279198, 'acc': 0.980469, time: 1.828
2022-09-15 16:15:40,799-INFO: epoch: 71, iter: 1140, lr: 0.001000, 'loss': 23.729788, 'acc': 0.988281, time: 0.374
2022-09-15 16:15:49,867-INFO: epoch: 72, iter: 1160, lr: 0.001000, 'loss': 19.145424, 'acc': 0.988281, time: 0.374
2022-09-15 16:15:58,914-INFO: epoch: 73, iter: 1180, lr: 0.001000, 'loss': 16.507511, 'acc': 0.992188, time: 0.374
2022-09-15 16:16:09,365-INFO: epoch: 75, iter: 1200, lr: 0.001000, 'loss': 19.851065, 'acc': 0.988281, time: 1.786
2022-09-15 16:16:18,561-INFO: epoch: 76, iter: 1220, lr: 0.001000, 'loss': 33.03997, 'acc': 0.980469, time: 0.404
2022-09-15 16:16:27,601-INFO: epoch: 77, iter: 1240, lr: 0.001000, 'loss': 43.73796, 'acc': 0.974609, time: 0.373
2022-09-15 16:16:36,731-INFO: epoch: 78, iter: 1260, lr: 0.001000, 'loss': 25.171104, 'acc': 0.984375, time: 0.374
2022-09-15 16:16:47,318-INFO: epoch: 80, iter: 1280, lr: 0.001000, 'loss': 19.046516, 'acc': 0.988281, time: 1.838
2022-09-15 16:16:56,417-INFO: epoch: 81, iter: 1300, lr: 0.001000, 'loss': 17.294367, 'acc': 0.988281, time: 0.399
2022-09-15 16:17:05,547-INFO: epoch: 82, iter: 1320, lr: 0.001000, 'loss': 17.236929, 'acc': 0.992188, time: 0.373
2022-09-15 16:17:14,582-INFO: epoch: 83, iter: 1340, lr: 0.001000, 'loss': 15.577, 'acc': 0.988281, time: 0.373
2022-09-15 16:17:25,033-INFO: epoch: 85, iter: 1360, lr: 0.001000, 'loss': 15.052889, 'acc': 0.990234, time: 1.760
2022-09-15 16:17:34,349-INFO: epoch: 86, iter: 1380, lr: 0.001000, 'loss': 17.350527, 'acc': 0.988281, time: 0.413
2022-09-15 16:17:43,599-INFO: epoch: 87, iter: 1400, lr: 0.001000, 'loss': 14.587466, 'acc': 0.990234, time: 0.374
2022-09-15 16:17:52,589-INFO: epoch: 88, iter: 1420, lr: 0.001000, 'loss': 14.150236, 'acc': 0.992188, time: 0.373
2022-09-15 16:18:02,967-INFO: epoch: 90, iter: 1440, lr: 0.001000, 'loss': 12.714052, 'acc': 0.988281, time: 1.742
2022-09-15 16:18:12,164-INFO: epoch: 91, iter: 1460, lr: 0.001000, 'loss': 10.972769, 'acc': 0.992188, time: 0.402
2022-09-15 16:18:21,412-INFO: epoch: 92, iter: 1480, lr: 0.001000, 'loss': 14.171873, 'acc': 0.992188, time: 0.374
2022-09-15 16:18:30,446-INFO: epoch: 93, iter: 1500, lr: 0.001000, 'loss': 10.255694, 'acc': 0.992188, time: 0.374
2022-09-15 16:18:32,318-INFO: Already save model in ./myoutput/rec_RARE_atten_new/best_accuracy
2022-09-15 16:18:32,319-INFO: Test iter: 1500, acc:0.823177, best_acc:0.823177, best_epoch:93, best_batch_id:1500, eval_sample_num:1001
2022-09-15 16:18:43,055-INFO: epoch: 95, iter: 1520, lr: 0.001000, 'loss': 9.569214, 'acc': 0.996094, time: 1.995
2022-09-15 16:18:52,264-INFO: epoch: 96, iter: 1540, lr: 0.001000, 'loss': 9.96125, 'acc': 0.992188, time: 0.414
2022-09-15 16:19:01,514-INFO: epoch: 97, iter: 1560, lr: 0.001000, 'loss': 8.359654, 'acc': 0.996094, time: 0.374
2022-09-15 16:19:10,616-INFO: epoch: 98, iter: 1580, lr: 0.001000, 'loss': 6.192225, 'acc': 0.998047, time: 0.373
2022-09-15 16:19:21,189-INFO: epoch: 100, iter: 1600, lr: 0.001000, 'loss': 5.406765, 'acc': 0.996094, time: 1.799
2022-09-15 16:19:27,310-INFO: Already save model in ./myoutput/rec_RARE_atten_new/iter_epoch_100
2022-09-15 16:19:30,699-INFO: epoch: 101, iter: 1620, lr: 0.001000, 'loss': 5.483102, 'acc': 0.996094, time: 0.381
2022-09-15 16:19:39,779-INFO: epoch: 102, iter: 1640, lr: 0.001000, 'loss': 9.586991, 'acc': 0.996094, time: 0.374
2022-09-15 16:19:48,841-INFO: epoch: 103, iter: 1660, lr: 0.001000, 'loss': 5.63769, 'acc': 0.996094, time: 0.374
2022-09-15 16:19:59,323-INFO: epoch: 105, iter: 1680, lr: 0.001000, 'loss': 11.556911, 'acc': 0.996094, time: 1.855
2022-09-15 16:20:08,442-INFO: epoch: 106, iter: 1700, lr: 0.001000, 'loss': 8.392323, 'acc': 0.996094, time: 0.420
2022-09-15 16:20:17,532-INFO: epoch: 107, iter: 1720, lr: 0.001000, 'loss': 4.550391, 'acc': 0.996094, time: 0.374
2022-09-15 16:20:26,593-INFO: epoch: 108, iter: 1740, lr: 0.001000, 'loss': 5.61542, 'acc': 0.996094, time: 0.373
2022-09-15 16:20:37,160-INFO: epoch: 110, iter: 1760, lr: 0.001000, 'loss': 9.220263, 'acc': 0.996094, time: 1.810
2022-09-15 16:20:46,332-INFO: epoch: 111, iter: 1780, lr: 0.001000, 'loss': 4.136215, 'acc': 0.998047, time: 0.422
2022-09-15 16:20:55,467-INFO: epoch: 112, iter: 1800, lr: 0.001000, 'loss': 5.135178, 'acc': 0.996094, time: 0.373
2022-09-15 16:21:04,534-INFO: epoch: 113, iter: 1820, lr: 0.001000, 'loss': 5.234288, 'acc': 0.996094, time: 0.374
2022-09-15 16:21:15,182-INFO: epoch: 115, iter: 1840, lr: 0.001000, 'loss': 4.517134, 'acc': 0.996094, time: 1.924
2022-09-15 16:21:24,325-INFO: epoch: 116, iter: 1860, lr: 0.001000, 'loss': 3.831606, 'acc': 1.0, time: 0.387
2022-09-15 16:21:33,438-INFO: epoch: 117, iter: 1880, lr: 0.001000, 'loss': 3.111336, 'acc': 0.996094, time: 0.373
2022-09-15 16:21:42,490-INFO: epoch: 118, iter: 1900, lr: 0.001000, 'loss': 3.794124, 'acc': 0.996094, time: 0.374
2022-09-15 16:21:53,098-INFO: epoch: 120, iter: 1920, lr: 0.001000, 'loss': 3.929718, 'acc': 0.998047, time: 1.913
2022-09-15 16:22:02,185-INFO: epoch: 121, iter: 1940, lr: 0.001000, 'loss': 4.611428, 'acc': 0.996094, time: 0.375
2022-09-15 16:22:11,267-INFO: epoch: 122, iter: 1960, lr: 0.001000, 'loss': 2.685765, 'acc': 1.0, time: 0.374
2022-09-15 16:22:20,275-INFO: epoch: 123, iter: 1980, lr: 0.001000, 'loss': 2.870256, 'acc': 1.0, time: 0.373
2022-09-15 16:22:30,732-INFO: epoch: 125, iter: 2000, lr: 0.001000, 'loss': 4.04394, 'acc': 0.998047, time: 1.891
2022-09-15 16:22:33,935-INFO: Already save model in ./myoutput/rec_RARE_atten_new/best_accuracy
2022-09-15 16:22:33,935-INFO: Test iter: 2000, acc:0.829171, best_acc:0.829171, best_epoch:125, best_batch_id:2000, eval_sample_num:1001
2022-09-15 16:22:42,925-INFO: epoch: 126, iter: 2020, lr: 0.001000, 'loss': 3.729886, 'acc': 0.996094, time: 0.384
2022-09-15 16:22:51,976-INFO: epoch: 127, iter: 2040, lr: 0.001000, 'loss': 3.797183, 'acc': 0.996094, time: 0.377
2022-09-15 16:23:01,064-INFO: epoch: 128, iter: 2060, lr: 0.001000, 'loss': 3.841343, 'acc': 1.0, time: 0.374
2022-09-15 16:23:11,632-INFO: epoch: 130, iter: 2080, lr: 0.001000, 'loss': 2.5157, 'acc': 1.0, time: 1.818
2022-09-15 16:23:20,885-INFO: epoch: 131, iter: 2100, lr: 0.001000, 'loss': 2.793916, 'acc': 1.0, time: 0.382
2022-09-15 16:23:29,893-INFO: epoch: 132, iter: 2120, lr: 0.001000, 'loss': 2.068662, 'acc': 1.0, time: 0.373
2022-09-15 16:23:38,892-INFO: epoch: 133, iter: 2140, lr: 0.001000, 'loss': 1.963152, 'acc': 1.0, time: 0.374
2022-09-15 16:23:49,438-INFO: epoch: 135, iter: 2160, lr: 0.001000, 'loss': 1.785324, 'acc': 1.0, time: 1.811
2022-09-15 16:23:58,622-INFO: epoch: 136, iter: 2180, lr: 0.001000, 'loss': 1.794544, 'acc': 1.0, time: 0.388
2022-09-15 16:24:07,731-INFO: epoch: 137, iter: 2200, lr: 0.001000, 'loss': 1.725564, 'acc': 1.0, time: 0.381
2022-09-15 16:24:16,818-INFO: epoch: 138, iter: 2220, lr: 0.001000, 'loss': 1.576424, 'acc': 1.0, time: 0.373
2022-09-15 16:24:27,274-INFO: epoch: 140, iter: 2240, lr: 0.001000, 'loss': 1.49527, 'acc': 1.0, time: 1.773
2022-09-15 16:24:36,422-INFO: epoch: 141, iter: 2260, lr: 0.001000, 'loss': 1.492407, 'acc': 1.0, time: 0.374
2022-09-15 16:24:45,728-INFO: epoch: 142, iter: 2280, lr: 0.001000, 'loss': 1.456, 'acc': 1.0, time: 0.374
2022-09-15 16:24:54,789-INFO: epoch: 143, iter: 2300, lr: 0.001000, 'loss': 1.191148, 'acc': 1.0, time: 0.373
2022-09-15 16:25:05,522-INFO: epoch: 145, iter: 2320, lr: 0.001000, 'loss': 1.210015, 'acc': 1.0, time: 1.863
2022-09-15 16:25:14,468-INFO: epoch: 146, iter: 2340, lr: 0.001000, 'loss': 1.281136, 'acc': 1.0, time: 0.375
2022-09-15 16:25:23,566-INFO: epoch: 147, iter: 2360, lr: 0.001000, 'loss': 1.303462, 'acc': 1.0, time: 0.374
2022-09-15 16:25:32,584-INFO: epoch: 148, iter: 2380, lr: 0.001000, 'loss': 1.070822, 'acc': 1.0, time: 0.375
2022-09-15 16:25:42,999-INFO: epoch: 150, iter: 2400, lr: 0.001000, 'loss': 1.107232, 'acc': 1.0, time: 1.789
2022-09-15 16:25:52,155-INFO: epoch: 151, iter: 2420, lr: 0.001000, 'loss': 1.304622, 'acc': 1.0, time: 0.411
2022-09-15 16:26:01,268-INFO: epoch: 152, iter: 2440, lr: 0.001000, 'loss': 1.403758, 'acc': 1.0, time: 0.374
2022-09-15 16:26:10,305-INFO: epoch: 153, iter: 2460, lr: 0.001000, 'loss': 1.210189, 'acc': 1.0, time: 0.373
2022-09-15 16:26:20,865-INFO: epoch: 155, iter: 2480, lr: 0.001000, 'loss': 1.069858, 'acc': 1.0, time: 1.815
2022-09-15 16:26:30,007-INFO: epoch: 156, iter: 2500, lr: 0.001000, 'loss': 1.320687, 'acc': 1.0, time: 0.375
2022-09-15 16:26:32,067-INFO: Already save model in ./myoutput/rec_RARE_atten_new/best_accuracy
2022-09-15 16:26:32,067-INFO: Test iter: 2500, acc:0.831169, best_acc:0.831169, best_epoch:156, best_batch_id:2500, eval_sample_num:1001
2022-09-15 16:26:41,097-INFO: epoch: 157, iter: 2520, lr: 0.001000, 'loss': 1.170897, 'acc': 1.0, time: 0.374
2022-09-15 16:26:50,126-INFO: epoch: 158, iter: 2540, lr: 0.001000, 'loss': 1.209216, 'acc': 1.0, time: 0.373
2022-09-15 16:27:00,639-INFO: epoch: 160, iter: 2560, lr: 0.001000, 'loss': 1.066662, 'acc': 1.0, time: 1.871
2022-09-15 16:27:09,739-INFO: epoch: 161, iter: 2580, lr: 0.001000, 'loss': 0.858585, 'acc': 1.0, time: 0.378
2022-09-15 16:27:18,841-INFO: epoch: 162, iter: 2600, lr: 0.001000, 'loss': 0.789717, 'acc': 1.0, time: 0.373
2022-09-15 16:27:27,905-INFO: epoch: 163, iter: 2620, lr: 0.001000, 'loss': 0.861278, 'acc': 1.0, time: 0.373
2022-09-15 16:27:38,476-INFO: epoch: 165, iter: 2640, lr: 0.001000, 'loss': 0.891585, 'acc': 1.0, time: 1.823
2022-09-15 16:27:47,624-INFO: epoch: 166, iter: 2660, lr: 0.001000, 'loss': 0.919595, 'acc': 1.0, time: 0.383
2022-09-15 16:27:56,740-INFO: epoch: 167, iter: 2680, lr: 0.001000, 'loss': 0.757441, 'acc': 1.0, time: 0.374
2022-09-15 16:28:05,765-INFO: epoch: 168, iter: 2700, lr: 0.001000, 'loss': 0.720211, 'acc': 1.0, time: 0.374
2022-09-15 16:28:16,255-INFO: epoch: 170, iter: 2720, lr: 0.001000, 'loss': 0.76394, 'acc': 1.0, time: 1.810
2022-09-15 16:28:25,394-INFO: epoch: 171, iter: 2740, lr: 0.001000, 'loss': 0.70124, 'acc': 1.0, time: 0.375
2022-09-15 16:28:34,568-INFO: epoch: 172, iter: 2760, lr: 0.001000, 'loss': 0.681117, 'acc': 1.0, time: 0.373
2022-09-15 16:28:43,565-INFO: epoch: 173, iter: 2780, lr: 0.001000, 'loss': 0.71791, 'acc': 1.0, time: 0.375
2022-09-15 16:28:54,132-INFO: epoch: 175, iter: 2800, lr: 0.001000, 'loss': 0.726069, 'acc': 1.0, time: 1.846
2022-09-15 16:29:03,266-INFO: epoch: 176, iter: 2820, lr: 0.001000, 'loss': 0.662951, 'acc': 1.0, time: 0.435
2022-09-15 16:29:12,259-INFO: epoch: 177, iter: 2840, lr: 0.001000, 'loss': 0.770077, 'acc': 1.0, time: 0.374
2022-09-15 16:29:21,314-INFO: epoch: 178, iter: 2860, lr: 0.001000, 'loss': 0.70409, 'acc': 1.0, time: 0.374
2022-09-15 16:29:31,740-INFO: epoch: 180, iter: 2880, lr: 0.001000, 'loss': 0.75167, 'acc': 1.0, time: 1.781
2022-09-15 16:29:40,952-INFO: epoch: 181, iter: 2900, lr: 0.001000, 'loss': 0.595518, 'acc': 1.0, time: 0.384
2022-09-15 16:29:50,018-INFO: epoch: 182, iter: 2920, lr: 0.001000, 'loss': 0.678415, 'acc': 1.0, time: 0.374
2022-09-15 16:29:59,274-INFO: epoch: 183, iter: 2940, lr: 0.001000, 'loss': 0.602751, 'acc': 1.0, time: 0.373
2022-09-15 16:30:09,946-INFO: epoch: 185, iter: 2960, lr: 0.001000, 'loss': 0.568185, 'acc': 1.0, time: 2.010
2022-09-15 16:30:19,154-INFO: epoch: 186, iter: 2980, lr: 0.001000, 'loss': 0.56914, 'acc': 1.0, time: 0.397
2022-09-15 16:30:28,234-INFO: epoch: 187, iter: 3000, lr: 0.001000, 'loss': 0.570147, 'acc': 1.0, time: 0.373
2022-09-15 16:30:29,647-INFO: Test iter: 3000, acc:0.831169, best_acc:0.831169, best_epoch:156, best_batch_id:2500, eval_sample_num:1001
2022-09-15 16:30:38,718-INFO: epoch: 188, iter: 3020, lr: 0.001000, 'loss': 0.556909, 'acc': 1.0, time: 0.373
2022-09-15 16:30:49,241-INFO: epoch: 190, iter: 3040, lr: 0.001000, 'loss': 0.522148, 'acc': 1.0, time: 1.889
2022-09-15 16:30:58,437-INFO: epoch: 191, iter: 3060, lr: 0.001000, 'loss': 0.498712, 'acc': 1.0, time: 0.404
2022-09-15 16:31:07,491-INFO: epoch: 192, iter: 3080, lr: 0.001000, 'loss': 0.428753, 'acc': 1.0, time: 0.373
2022-09-15 16:31:16,654-INFO: epoch: 193, iter: 3100, lr: 0.001000, 'loss': 0.434947, 'acc': 1.0, time: 0.374
2022-09-15 16:31:27,201-INFO: epoch: 195, iter: 3120, lr: 0.001000, 'loss': 0.485586, 'acc': 1.0, time: 1.832
2022-09-15 16:31:36,403-INFO: epoch: 196, iter: 3140, lr: 0.001000, 'loss': 0.46684, 'acc': 1.0, time: 0.392
2022-09-15 16:31:45,477-INFO: epoch: 197, iter: 3160, lr: 0.001000, 'loss': 0.513497, 'acc': 1.0, time: 0.374
2022-09-15 16:31:54,520-INFO: epoch: 198, iter: 3180, lr: 0.001000, 'loss': 0.657723, 'acc': 1.0, time: 0.374

导出推理模型

In [78]

%cd ~/PaddleOCR
# 导出检测模型
!python3 tools/export_model.py \\
        -c configs/det/det_mv3_db.yml \\
        -o Global.checkpoints=./myoutput/det_db/best_accuracy \\
        Global.save_inference_dir=./inference/mydet_db

# 导出识别模型
!python3 tools/export_model.py \\
        -c configs/rec/rec_mv3_tps_bilstm_attn.yml \\
        -o Global.checkpoints=./myoutput/rec_RARE_atten_new/best_accuracy \\
        Global.save_inference_dir=./inference/myrec_rare

/home/aistudio/work/PaddleOCR
2022-09-15 16:37:11,326-INFO: 'Global': 'debug': False, 'algorithm': 'DB', 'use_gpu': True, 'epoch_num': 1200, 'log_smooth_window': 20, 'print_batch_step': 20, 'save_model_dir': './myoutput/det_db/', 'save_epoch_step': 10, 'eval_batch_step': [100, 500], 'train_batch_size_per_card': 4, 'test_batch_size_per_card': 4, 'image_shape': [3, 640, 640], 'reader_yml': './configs/det/det_db_icdar15_reader.yml', 'pretrain_weights': './pretrain_models/det_mv3_db/best_accuracy', 'checkpoints': './myoutput/det_db/best_accuracy', 'save_res_path': './myoutput/det_db/predicts_db.txt', 'save_inference_dir': './inference/mydet_db', 'Architecture': 'function': 'ppocr.modeling.architectures.det_model,DetModel', 'Backbone': 'function': 'ppocr.modeling.backbones.det_mobilenet_v3,MobileNetV3', 'scale': 0.5, 'model_name': 'large', 'Head': 'function': 'ppocr.modeling.heads.det_db_head,DBHead', 'model_name': 'large', 'k': 50, 'inner_channels': 96, 'out_channels': 2, 'Loss': 'function': 'ppocr.modeling.losses.det_db_loss,DBLoss', 'balance_loss': True, 'main_loss_type': 'DiceLoss', 'alpha': 5, 'beta': 10, 'ohem_ratio': 3, 'Optimizer': 'function': 'ppocr.optimizer,AdamDecay', 'base_lr': 0.001, 'beta1': 0.9, 'beta2': 0.999, 'PostProcess': 'function': 'ppocr.postprocess.db_postprocess,DBPostProcess', 'thresh': 0.3, 'box_thresh': 0.7, 'max_candidates': 1000, 'unclip_ratio': 2.0, 'TrainReader': 'reader_function': 'ppocr.data.det.dataset_traversal,TrainReader', 'process_function': 'ppocr.data.det.db_process,DBProcessTrain', 'num_workers': 1, 'img_set_dir': '../../', 'label_file_path': '../train_data_det.txt', 'EvalReader': 'reader_function': 'ppocr.data.det.dataset_traversal,EvalTestReader', 'process_function': 'ppocr.data.det.db_process,DBProcessTest', 'img_set_dir': '../../', 'label_file_path': '../val_data_det.txt', 'test_image_shape': [736, 1280], 'TestReader': 'reader_function': 'ppocr.data.det.dataset_traversal,EvalTestReader', 'process_function': 'ppocr.data.det.db_process,DBProcessTest', 'infer_img': None, 'img_set_dir': '../../', 'label_file_path': '../test_data_det.txt', 'test_image_shape': [736, 1280], 'do_eval': True
3 640 640
W0915 16:37:11.688948 25074 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 11.2, Runtime API Version: 9.0
W0915 16:37:11.693445 25074 device_context.cc:260] device: 0, cuDNN Version: 7.6.
2022-09-15 16:37:13,747-INFO: Finish initing model from ./myoutput/det_db/best_accuracy
inference model saved in ./inference/mydet_db/model and ./inference/mydet_db/params
save success, output_name_list: ['maps']
2022-09-15 16:37:16,325-INFO: 'Global': 'debug': False, 'algorithm': 'RARE', 'use_gpu': True, 'epoch_num': 200, 'log_smooth_window': 20, 'print_batch_step': 20, 'save_model_dir': './myoutput/rec_RARE_atten_new', 'save_epoch_step': 100, 'eval_batch_step': 500, 'train_batch_size_per_card': 256, 'test_batch_size_per_card': 256, 'image_shape': [3, 32, 320], 'max_text_length': 8, 'character_type': 'ch', 'character_dict_path': '../word_dict.txt', 'loss_type': 'attention', 'tps': True, 'reader_yml': './configs/rec/rec_chinese_reader.yml', 'pretrain_weights': './pretrain_models/rec_mv3_tps_bilstm_attn/best_accuracy', 'checkpoints': './myoutput/rec_RARE_atten_new/best_accuracy', 'save_inference_dir': './inference/myrec_rare', 'infer_img': None, 'Architecture': 'function': 'ppocr.modeling.architectures.rec_model,RecModel', 'TPS': 'function': 'ppocr.modeling.stns.tps,TPS', 'num_fiducial': 20, 'loc_lr': 0.1, 'model_name': 'small', 'Backbone': 'function': 'ppocr.modeling.backbones.rec_mobilenet_v3,MobileNetV3', 'scale': 0.5, 'model_name': 'large', 'Head': 'function': 'ppocr.modeling.heads.rec_attention_head,AttentionPredict', 'encoder_type': 'rnn', 'SeqRNN': 'hidden_size': 96, 'Attention': 'decoder_size': 96, 'word_vector_dim': 96, 'Loss': 'function': 'ppocr.modeling.losses.rec_attention_loss,AttentionLoss', 'Optimizer': 'function': 'ppocr.optimizer,AdamDecay', 'base_lr': 0.001, 'beta1': 0.9, 'beta2': 0.999, 'TrainReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader', 'num_workers': 8, 'img_set_dir': '../..', 'label_file_path': '../train_data_rec.txt', 'EvalReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader', 'num_workers': 8, 'img_set_dir': '../..', 'label_file_path': '../val_data_rec.txt', 'TestReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader'
W0915 16:37:16.958099 25152 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 11.2, Runtime API Version: 9.0
W0915 16:37:16.962687 25152 device_context.cc:260] device: 0, cuDNN Version: 7.6.
2022-09-15 16:37:19,039-INFO: Finish initing model from ./myoutput/rec_RARE_atten_new/best_accuracy
inference model saved in ./inference/myrec_rare/model and ./inference/myrec_rare/params
save success, output_name_list: ['decoded_out', 'predicts']

模型测试

In [79]

%cd ~/PaddleOCR
!python3 tools/infer/predict_system.py \\
    --image_dir="../../imgtest" \\
    --det_model_dir="./inference/mydet_db" \\
    --rec_model_dir="./inference/myrec_rare" \\
    --rec_image_shape="3, 32, 320" \\
    --rec_char_type="ch" \\
    --rec_algorithm="RARE" \\
    --use_space_char False \\
    --max_text_length 8 \\
    --rec_char_dict_path="../word_dict.txt" \\
    --use_gpu False

/home/aistudio/work/PaddleOCR
dt_boxes num : 1, elapse : 0.9093203544616699
rec_res num  : 1, elapse : 0.051482200622558594
Predict time of ../../imgtest/0425-88_255-205&515_553&627-528&596_205&627_224&532_553&515-0_0_3_25_27_30_33_33-72-8.jpg: 0.981s
皖AD13699, 1.000
The visualized image saved in ./inference_results/0425-88_255-205&515_553&627-528&596_205&627_224&532_553&515-0_0_3_25_27_30_33_33-72-8.jpg
dt_boxes num : 1, elapse : 0.6962370872497559
rec_res num  : 1, elapse : 0.046535491943359375
Predict time of ../../imgtest/04-90_267-207&511_555&616-550&605_211&616_207&515_555&511-0_0_3_1_26_30_31_30-68-81.jpg: 0.754s
皖AD82676, 0.995
The visualized image saved in ./inference_results/04-90_267-207&511_555&616-550&605_211&616_207&515_555&511-0_0_3_1_26_30_31_30-68-81.jpg
dt_boxes num : 1, elapse : 0.6199753284454346
rec_res num  : 1, elapse : 0.08345222473144531
Predict time of ../../imgtest/0375-92_263-192&480_552&575-536&575_198&560_192&480_552&495-0_0_5_24_29_33_24_24-133-69.jpg: 0.746s
皖AF05900, 1.000
The visualized image saved in ./inference_results/0375-92_263-192&480_552&575-536&575_198&560_192&480_552&495-0_0_5_24_29_33_24_24-133-69.jpg
dt_boxes num : 1, elapse : 0.6378555297851562
rec_res num  : 1, elapse : 0.08243370056152344
Predict time of ../../imgtest/0475-68_277-140&493_368&683-368&570_162&683_140&590_350&493-0_0_5_25_29_32_32_30-170-444.jpg: 0.733s
皖AF15886, 0.999
The visualized image saved in ./inference_results/0475-68_277-140&493_368&683-368&570_162&683_140&590_350&493-0_0_5_25_29_32_32_30-170-444.jpg
dt_boxes num : 1, elapse : 0.6511027812957764
rec_res num  : 1, elapse : 0.04664444923400879
Predict time of ../../imgtest/05-90_257-137&507_572&612-567&611_158&612_137&507_572&512-0_0_3_28_32_25_24_32-144-135.jpg: 0.709s
皖AD48108, 0.999
The visualized image saved in ./inference_results/05-90_257-137&507_572&612-567&611_158&612_137&507_572&512-0_0_3_28_32_25_24_32-144-135.jpg
dt_boxes num : 1, elapse : 0.6640639305114746
rec_res num  : 1, elapse : 0.08099365234375
Predict time of ../../imgtest/03-103_253-267&425_483&565-483&565_271&497_267&425_480&483-0_0_3_25_25_33_25_25-110-47.jpg: 0.756s
皖AD11911, 1.000
The visualized image saved in ./inference_results/03-103_253-267&425_483&565-483&565_271&497_267&425_480&483-0_0_3_25_25_33_25_25-110-47.jpg
dt_boxes num : 1, elapse : 0.6204874515533447
rec_res num  : 1, elapse : 0.08555459976196289
Predict time of ../../imgtest/0375-90_256-181&548_541&643-541&643_200&633_181&548_529&553-0_0_3_2_29_33_33_26-176-389.jpg: 0.717s
皖AD05992, 0.919
The visualized image saved in ./inference_results/0375-90_256-181&548_541&643-541&643_200&633_181&548_529&553-0_0_3_2_29_33_33_26-176-389.jpg
dt_boxes num : 1, elapse : 0.613457441329956
rec_res num  : 1, elapse : 0.08662247657775879
Predict time of ../../imgtest/0475-90_238-143&505_599&600-599&600_186&586_143&509_577&505-0_0_3_30_25_27_27_32-60-74.jpg: 0.711s
皖AD61338, 0.999
The visualized image saved in ./inference_results/0475-90_238-143&505_599&600-599&600_186&586_143&509_577&505-0_0_3_30_25_27_27_32-60-74.jpg
dt_boxes num : 1, elapse : 0.6121721267700195
rec_res num  : 1, elapse : 0.08241701126098633
Predict time of ../../imgtest/01-90_265-231&522_405&574-405&571_235&574_231&523_403&522-0_0_3_1_28_29_30_30-134-56.jpg: 0.705s
皖AD84566, 0.996
The visualized image saved in ./inference_results/01-90_265-231&522_405&574-405&571_235&574_231&523_403&522-0_0_3_1_28_29_30_30-134-56.jpg
dt_boxes num : 1, elapse : 0.6165649890899658
rec_res num  : 1, elapse : 0.0794670581817627
Predict time of ../../imgtest/04-91_254-145&472_529&567-529&567_164&552_145&472_529&485-0_0_3_25_24_24_24_29-148-355.jpg: 0.707s
皖AD10005, 1.000
The visualized image saved in ./inference_results/04-91_254-145&472_529&567-529&567_164&552_145&472_529&485-0_0_3_25_24_24_24_29-148-355.jpg

In [80]

!python tools/eval.py -c  configs/rec/rec_mv3_tps_bilstm_attn.yml \\
-o Global.checkpoints=./myoutput/rec_RARE_atten_new/best_accuracy

2022-09-15 16:39:43,444-INFO: 'Global': 'debug': False, 'algorithm': 'RARE', 'use_gpu': True, 'epoch_num': 200, 'log_smooth_window': 20, 'print_batch_step': 20, 'save_model_dir': './myoutput/rec_RARE_atten_new', 'save_epoch_step': 100, 'eval_batch_step': 500, 'train_batch_size_per_card': 256, 'test_batch_size_per_card': 256, 'image_shape': [3, 32, 320], 'max_text_length': 8, 'character_type': 'ch', 'character_dict_path': '../word_dict.txt', 'loss_type': 'attention', 'tps': True, 'reader_yml': './configs/rec/rec_chinese_reader.yml', 'pretrain_weights': './pretrain_models/rec_mv3_tps_bilstm_attn/best_accuracy', 'checkpoints': './myoutput/rec_RARE_atten_new/best_accuracy', 'save_inference_dir': './inference/newimg_rec_rare', 'infer_img': None, 'Architecture': 'function': 'ppocr.modeling.architectures.rec_model,RecModel', 'TPS': 'function': 'ppocr.modeling.stns.tps,TPS', 'num_fiducial': 20, 'loc_lr': 0.1, 'model_name': 'small', 'Backbone': 'function': 'ppocr.modeling.backbones.rec_mobilenet_v3,MobileNetV3', 'scale': 0.5, 'model_name': 'large', 'Head': 'function': 'ppocr.modeling.heads.rec_attention_head,AttentionPredict', 'encoder_type': 'rnn', 'SeqRNN': 'hidden_size': 96, 'Attention': 'decoder_size': 96, 'word_vector_dim': 96, 'Loss': 'function': 'ppocr.modeling.losses.rec_attention_loss,AttentionLoss', 'Optimizer': 'function': 'ppocr.optimizer,AdamDecay', 'base_lr': 0.001, 'beta1': 0.9, 'beta2': 0.999, 'TrainReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader', 'num_workers': 8, 'img_set_dir': '../..', 'label_file_path': '../train_data_rec.txt', 'EvalReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader', 'num_workers': 8, 'img_set_dir': '../..', 'label_file_path': '../val_data_rec.txt', 'TestReader': 'reader_function': 'ppocr.data.rec.dataset_traversal,SimpleReader'
W0915 16:39:44.065701 25553 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 11.2, Runtime API Version: 9.0
W0915 16:39:44.070394 25553 device_context.cc:260] device: 0, cuDNN Version: 7.6.
2022-09-15 16:39:46,158-INFO: Finish initing model from ./myoutput/rec_RARE_atten_new/best_accuracy
2022-09-15 16:39:47,598-INFO: Eval result: 'avg_acc': 0.8311688311688312, 'total_acc_num': 832, 'total_sample_num': 1001

测试集部分识别结果展示

总结：

本项目基于Paddleocr完成深度学习车牌识别，项目分为车牌检测与车牌识别两部分，最终完整实现车牌识别目的，可以识别大部分车牌结果。针对本项目的识别准确度，后续优化可以增加数据增强方面内容、更换其他主干网络及对应参数调优，大家可以在此项目基础上进行扩充，欢迎评论区一起讨论！

作者博客：CSDN主页（专注大数据与人工智能知识分享，欢迎关注！）

以上是关于Paddle入门实战系列：基于PaddleOCR的车牌识别的主要内容，如果未能解决你的问题，请参考以下文章

Paddle入门实战系列：车牌检测与识别

Paddle入门实战系列：中文场景文字识别

paddleocr使用问题01_segment_fatal

Paddle进阶实战系列：智慧交通预测系统