PaddleOCR车牌检测识别训练、部署
需要的文件 1) CCPD2019 数据集。2) 安装paddleOCR 和paddle 等环境。 3)本机显卡3080，cuda 11.0 cudnn 8.0.5 paddle 2.0.2 paddleocr 2.0.6。
- 下载数据集：
ccpd中科大数据集
数据集
- 本次使用的数据集为CCPD2019车牌数据集
- 该数据集在合肥市的停车场采集得来，采集时间早上7:30到晚上10:00。停车场采集人员手持Android POS机对停车场的车辆拍照并手工标注车牌位置。拍摄的车牌照片涉及多种复杂环境，包括模糊、倾斜、阴雨天、雪天等等。CCPD数据集一共包含将近30万张图片，每种图片大小720x1160x3。一共包含8项，具体如下：
- 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…
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’]
# 解压数据集
CCPD2019.zip --> CCPD
- 数据转化,生成数据列表
import os, cv2
import random
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
total = []
# E:\Paddle\PaddleOCR-release-2.1\work\CCPD
paths = os.listdir('E:\Paddle\PaddleOCR-release-2.1\work\CCPD')
for path in paths:
for item in os.listdir(os.path.join('E:\Paddle\PaddleOCR-release-2.1\work\CCPD', path)):
new_path = os.path.join(path, 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 = new_path+'\t'+'[{"transcription": "%s", "points": %s}]' % (' ', str(points))
line = line[:]+'\n'
total.append(line)
print(len(total))
random.shuffle(total)
with open('E:\Paddle\PaddleOCR-release-2.1\work\\train.txt', 'w', encoding='UTF-8') as f:
for line in total[:-1431]:
f.write(line)
with open('E:\Paddle\PaddleOCR-release-2.1\work\dev.txt', 'w', encoding='UTF-8') as f:
for line in total[-1431:]:
f.write(line)
- 下载检测和识别预训练模型
检测预训练模型
https://paddle-imagenet-models-name.bj.bcebos.com/MobileNetV3_large_x0_5_pretrained.tar
识别推理模型
https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar
- 检测模型训练
python tools/train.py -c det.yml
- det.yml文件
Global:
use_gpu: true
epoch_num: 5
log_smooth_window: 20
print_batch_step: 10
save_model_dir: output/det_vlpr/
save_epoch_step: 1
eval_batch_step: [1000, 1000]
load_static_weights: True
cal_metric_during_train: False
pretrained_model: pretrain_models/MobileNetV3_large_x0_5_pretrained
checkpoints:
save_inference_dir: inference/det_vlpr
use_visualdl: False
infer_img: doc/imgs_en/img_10.jpg
save_res_path: output/det_vlpr/predicts.txt
Architecture:
model_type: det
algorithm: DB
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: large
disable_se: True
Neck:
name: DBFPN
out_channels: 96
Head:
name: DBHead
k: 50
Loss:
name: DBLoss
balance_loss: true
main_loss_type: DiceLoss
alpha: 5
beta: 10
ohem_ratio: 3
Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
name: Cosine
learning_rate: 0.001
warmup_epoch: 1
regularizer:
name: 'L2'
factor: 0
PostProcess:
name: DBPostProcess
thresh: 0.3
box_thresh: 0.6
max_candidates: 1000
unclip_ratio: 1.5
Metric:
name: DetMetric
main_indicator: hmean
Train:
dataset:
name: SimpleDataSet
data_dir: CCPD
label_file_list:
- CCPD/train.txt
ratio_list: [1.0]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
- IaaAugment:
augmenter_args:
- {
'type': Fliplr, 'args': {
'p': 0.5 } }
- {
'type': Affine, 'args': {
'rotate': [-10, 10] } }
- {
'type': Resize, 'args': {
'size': [0.5, 3] } }
- EastRandomCropData:
size: [960, 960]
max_tries: 50
keep_ratio: true
- MakeBorderMap:
shrink_ratio: 0.4
thresh_min: 0.3
thresh_max: 0.7
- MakeShrinkMap:
shrink_ratio: 0.4
min_text_size: 8
- NormalizeImage:
scale: 1./255.
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
loader:
shuffle: True
drop_last: False
batch_size_per_card: 32
num_workers: 0
Eval:
dataset:
name: SimpleDataSet
data_dir: CCPD
label_file_list:
- CCPD/dev.txt
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
- DetResizeForTest:
- NormalizeImage:
scale: 1./255.
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
loader:
shuffle: False
drop_last: False
batch_size_per_card: 1 # must be 1
num_workers: 0
- 模型测试和模型导出
# 模型测试
python tools/infer_det.py -c det_vlpr.yml -o Global.infer_img=test0.jpg Global.pretrained_model=output/det_vlpr/best_accuracy Global.load_static_weights=False
# 模型导出
python tools/export_model.py -c output/det_vlpr/config.yml -o Global.pretrained_model=output/det_vlpr/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./inference/det_vlpr
- 模型部署
