网站建设中 显示,电子商务网页制作是什么,网站建设维修服务流程,长春关键词搜索排名IA-YOLO项目源自论文Image-Adaptive YOLO for Object Detection in Adverse Weather Conditions#xff0c;其提出端到端方式联合学习CNN-PP和YOLOv3#xff0c;这确保了CNN-PP可以学习适当的DIP#xff0c;以弱监督的方式增强图像检测。IA-YOLO方法可以自适应地处理正常和不…IA-YOLO项目源自论文Image-Adaptive YOLO for Object Detection in Adverse Weather Conditions其提出端到端方式联合学习CNN-PP和YOLOv3这确保了CNN-PP可以学习适当的DIP以弱监督的方式增强图像检测。IA-YOLO方法可以自适应地处理正常和不利天气条件下的图像。阅读论文发现只介绍了IA-YOLO的实际效果对DIP模块和CNN-PP介绍较少故此查阅源码分析其实现。
通过分析发现IA-YOLO使用filtered_image_batch 与 input_data_clean的mse loss优化CNN-PP模块的输出。故此可以看出CNN-PP | DIP实则是可以从IA-YOLO中剥离出来单独作为数据优化模块使用若需要考虑IA-YOLO项目的使用应该对比研究其它图像增强模块。
1、IA-YOLO项目的使用
1.1、安装命令
$ git clone https://github.com/wenyyu/Image-Adaptive-YOLO.git
$ cd Image-Adaptive-YOLO
# Require python3 and tensorflow
$ pip install -r ./docs/requirements.txt1.2、相关数据集
以下两个数据集是IA-YOLO使用到的外部数据集其还使用了voc数据集和foggy_voc数据集。 ExDark : https://github.com/cs-chan/Exclusively-Dark-Image-Dataset/tree/master/Dataset
RTTS: https://sites.google.com/view/reside-dehaze-datasets/reside-%CE%B2
1.3 基本使用
Train and Evaluate on the datasets
Download VOC PASCAL trainval and test data
$ wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
$ wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
$ wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tarExtract all of these tars into one directory and rename them, which should have the following basic structure. VOC # path: /home/lwy/work/code/tensorflow-yolov3/data/VOC
├── test
| └──VOCdevkit
| └──VOC2007 (from VOCtest_06-Nov-2007.tar)
└── train└──VOCdevkit└──VOC2007 (from VOCtrainval_06-Nov-2007.tar)└──VOC2012 (from VOCtrainval_11-May-2012.tar)$ python scripts/voc_annotation.pyGenerate Voc_foggy_train and Voc_foggy_val dataset offline
# generate ten levels foggy training images and val images, respectively
$ python ./core/data_make.py Edit core/config.py to configure
--vocfog_traindata_dir /data/vdd/liuwenyu/data_vocfog/train/JPEGImages/
--vocfog_valdata_dir /data/vdd/liuwenyu/data_vocfog/val/JPEGImages/
--train_path ./data/dataset_fog/voc_norm_train.txt
--test_path ./data/dataset_fog/voc_norm_test.txt
--class_name ./data/classes/vocfog.namesTrain and Evaluate
$ python train.py # we trained our model from scratch.
$ python evaluate.py
$ cd ./experiments/.../mAP python main.py More details of Preparing dataset or Train with your own dataset reference the implementation tensorflow-yolov3.
2、CNN-PP与DIP模块实现
在IA-YOLO论文中提到了不少数据处理知识如生成带雾图片、可微的DIP模块、CNN-PP模块等。这里主要介绍可微的DIP模块、CNN-PP模块的实现与训练
2.1 主体流程
如论文中所述缩略图与原始图的清晰化参数是一样的为节省计算量CNN-PP输入的是低分辨率的原始图像其输出的参数用于DIP模块进行图像清晰化而DIP模块则使用CNN-PP输出的参数优化图像最后交给yolov3模型进行预测
2.2 主体代码
为了便于进行对比实验IA-YOLO通过在yolov3.py(包含去雾filter)或yolov3_lowlight.py(不包含去雾filter,对图像随机进行亮度处理)中使用isp_flag作为配置项目决定是否使用CNN-PP与DIP模块。代码示意如下
#代码地址https://github.com/wenyyu/Image-Adaptive-YOLO/blob/main/core/yolov3_lowlight.pydef __build_nework(self, input_data, isp_flag, input_data_clean):filtered_image_batch input_dataself.filter_params input_datafilter_imgs_series []if isp_flag:with tf.variable_scope(extract_parameters_2):input_data tf.image.resize_images(input_data, [256, 256], methodtf.image.ResizeMethod.BILINEAR)#---------对原始图像进行下采样filter_features common.extract_parameters_2(input_data, cfg, self.trainable)#-----CNN-PP计算出DIP模块的参数# filter_features tf.random_normal([1, 10], 0.5, 0.1)filters cfg.filtersfilters [x(input_data, cfg) for x in filters]#-----生成DIP模块filter_parameters []for j, filter in enumerate(filters):with tf.variable_scope(filter_%d % j):print( creating filter:, j, name:, str(filter.__class__), abbr.,filter.get_short_name())print( filter_features:, filter_features.shape)filtered_image_batch, filter_parameter filter.apply(filtered_image_batch, filter_features)#-----DIP模块中filter使用CNN-PP参数优化图像filter_parameters.append(filter_parameter)filter_imgs_series.append(filtered_image_batch)print( output:, filtered_image_batch.shape)self.filter_params filter_parametersself.image_isped filtered_image_batchself.filter_imgs_series filter_imgs_seriesrecovery_loss tf.reduce_sum(tf.pow(filtered_image_batch - input_data_clean, 2.0))#/(2.0 * batch_size)#正常的yolov3代码训练使用流程input_data filtered_image_batch在if isp_flag内的代码简单介绍了CNN-DIP模块和DIP模块的使用 在if isp_flag外的代码为正常的yolov3训练代码。
2.3 CNN-PP模块的实现
完整代码https://github.com/wenyyu/Image-Adaptive-YOLO/blob/main/core/common.py
def extract_parameters_2(net, cfg, trainable):output_dim cfg.num_filter_parameters# net net - 0.5min_feature_map_size 4print(extract_parameters_2 CNN:)channels 16print( , str(net.get_shape()))net convolutional(net, filters_shape(3, 3, 3, channels), trainabletrainable, nameex_conv0,downsampleTrue, activateTrue, bnFalse)net convolutional(net, filters_shape(3, 3, channels, 2*channels), trainabletrainable, nameex_conv1,downsampleTrue, activateTrue, bnFalse)net convolutional(net, filters_shape(3, 3, 2*channels, 2*channels), trainabletrainable, nameex_conv2,downsampleTrue, activateTrue, bnFalse)net convolutional(net, filters_shape(3, 3, 2*channels, 2*channels), trainabletrainable, nameex_conv3,downsampleTrue, activateTrue, bnFalse)net convolutional(net, filters_shape(3, 3, 2*channels, 2*channels), trainabletrainable, nameex_conv4,downsampleTrue, activateTrue, bnFalse)net tf.reshape(net, [-1, 2048])features ly.fully_connected(net,64,scopefc1,activation_fnlrelu,weights_initializertf.contrib.layers.xavier_initializer())filter_features ly.fully_connected(features,output_dim,scopefc2,activation_fnNone,weights_initializertf.contrib.layers.xavier_initializer())return filter_features查阅代码可预见CNN-DIP为一个普通的CNN网络其输出结构由cfg.num_filter_parameters所决定具体值为14(不包含去雾filter)或15(包含去雾filter) 此外在config.py中定义了一些其他DIP、CNN-PP参数项
cfg.filters [DefogFilter, ImprovedWhiteBalanceFilter, GammaFilter,ToneFilter, ContrastFilter, UsmFilter
]
cfg.num_filter_parameters 15cfg.defog_begin_param 0cfg.wb_begin_param 1
cfg.gamma_begin_param 4
cfg.tone_begin_param 5
cfg.contrast_begin_param 13
cfg.usm_begin_param 14cfg.curve_steps 8
cfg.gamma_range 3
cfg.exposure_range 3.5
cfg.wb_range 1.1
cfg.color_curve_range (0.90, 1.10)
cfg.lab_curve_range (0.90, 1.10)
cfg.tone_curve_range (0.5, 2)
cfg.defog_range (0.1, 1.0)
cfg.usm_range (0.0, 5)# Masking is DISABLED
cfg.masking False
cfg.minimum_strength 0.3
cfg.maximum_sharpness 1
cfg.clamp False###########################################################################
# CNN Parameters
###########################################################################
cfg.source_img_size 64
cfg.base_channels 32
cfg.dropout_keep_prob 0.5
# G and C use the same feed dict?
cfg.share_feed_dict True
cfg.shared_feature_extractor True
cfg.fc1_size 128
cfg.bnw False
# number of filters for the first convolutional layers for all networks
# (stochastic/deterministic policy, critic, value)
cfg.feature_extractor_dims 40962.4 DIP模块的实现
DIP模块实质为堆叠的可微分滤波器其实现功能包括 Defog、白平衡WB、伽玛、对比度、Tone 和Sharpen其是实现代码主要在filters.py中部分辅助代码在util_filters.py中 在作者实现中有filters.py和filters_lowlight.py二者没有本质区别filters.py为包含去雾参数的filer代码中有部分注释未删除而filters_lowlight.py为后期更新代码实现了不包含去雾参数的filer删除了原有的注释
filters.py实现
下面代码中部分用到了tf的api来处理图像但其所有的Filter子类均非tf.layer对象使用tf——api处理图像只为了实现对CNN-PP输出结果的可微分
import tensorflow as tf
import numpy as np
import tensorflow.contrib.layers as ly
from util_filters import lrelu, rgb2lum, tanh_range, lerp
import cv2
import math
class Filter:def __init__(self, net, cfg):self.cfg cfg# self.height, self.width, self.channels list(map(int, net.get_shape()[1:]))# Specified in child classesself.num_filter_parameters Noneself.short_name Noneself.filter_parameters Nonedef get_short_name(self):assert self.short_namereturn self.short_namedef get_num_filter_parameters(self):assert self.num_filter_parametersreturn self.num_filter_parametersdef get_begin_filter_parameter(self):return self.begin_filter_parameterdef extract_parameters(self, features):# output_dim self.get_num_filter_parameters(# ) self.get_num_mask_parameters()# features ly.fully_connected(# features,# self.cfg.fc1_size,# scopefc1,# activation_fnlrelu,# weights_initializertf.contrib.layers.xavier_initializer())# features ly.fully_connected(# features,# output_dim,# scopefc2,# activation_fnNone,# weights_initializertf.contrib.layers.xavier_initializer())return features[:, self.get_begin_filter_parameter():(self.get_begin_filter_parameter() self.get_num_filter_parameters())], \features[:, self.get_begin_filter_parameter():(self.get_begin_filter_parameter() self.get_num_filter_parameters())]# Should be implemented in child classesdef filter_param_regressor(self, features):assert False# Process the whole image, without masking# Should be implemented in child classesdef process(self, img, param, defog, IcA):assert Falsedef debug_info_batched(self):return Falsedef no_high_res(self):return False# Apply the whole filter with maskingdef apply(self,img,img_featuresNone,defog_ANone,IcANone,specified_parameterNone,high_resNone):assert (img_features is None) ^ (specified_parameter is None)if img_features is not None:filter_features, mask_parameters self.extract_parameters(img_features)filter_parameters self.filter_param_regressor(filter_features)else:assert not self.use_masking()filter_parameters specified_parametermask_parameters tf.zeros(shape(1, self.get_num_mask_parameters()), dtypenp.float32)if high_res is not None:# working on high res...passdebug_info {}# We only debug the first image of this batchif self.debug_info_batched():debug_info[filter_parameters] filter_parameterselse:debug_info[filter_parameters] filter_parameters[0]# self.mask_parameters mask_parameters# self.mask self.get_mask(img, mask_parameters)# debug_info[mask] self.mask[0]#low_res_output lerp(img, self.process(img, filter_parameters), self.mask)low_res_output self.process(img, filter_parameters, defog_A, IcA)if high_res is not None:if self.no_high_res():high_res_output high_reselse:self.high_res_mask self.get_mask(high_res, mask_parameters)# high_res_output lerp(high_res,# self.process(high_res, filter_parameters, defog, IcA),# self.high_res_mask)else:high_res_output None#return low_res_output, high_res_output, debug_inforeturn low_res_output, filter_parametersdef use_masking(self):return self.cfg.maskingdef get_num_mask_parameters(self):return 6# Input: no need for tanh or sigmoid# Closer to 1 values are applied by filter more strongly# no additional TF variables insidedef get_mask(self, img, mask_parameters):if not self.use_masking():print(* Masking Disabled)return tf.ones(shape(1, 1, 1, 1), dtypetf.float32)else:print(* Masking Enabled)with tf.name_scope(namemask):# Six parameters for one filterfilter_input_range 5assert mask_parameters.shape[1] self.get_num_mask_parameters()mask_parameters tanh_range(l-filter_input_range, rfilter_input_range,initial0)(mask_parameters)size list(map(int, img.shape[1:3]))grid np.zeros(shape[1] size [2], dtypenp.float32)shorter_edge min(size[0], size[1])for i in range(size[0]):for j in range(size[1]):grid[0, i, j,0] (i (shorter_edge - size[0]) / 2.0) / shorter_edge - 0.5grid[0, i, j,1] (j (shorter_edge - size[1]) / 2.0) / shorter_edge - 0.5grid tf.constant(grid)# Ax By C * L Dinp grid[:, :, :, 0, None] * mask_parameters[:, None, None, 0, None] \grid[:, :, :, 1, None] * mask_parameters[:, None, None, 1, None] \mask_parameters[:, None, None, 2, None] * (rgb2lum(img) - 0.5) \mask_parameters[:, None, None, 3, None] * 2# Sharpness and inversioninp * self.cfg.maximum_sharpness * mask_parameters[:, None, None, 4,None] / filter_input_rangemask tf.sigmoid(inp)# Strengthmask mask * (mask_parameters[:, None, None, 5, None] / filter_input_range * 0.5 0.5) * (1 - self.cfg.minimum_strength) self.cfg.minimum_strengthprint(mask, mask.shape)return mask# def visualize_filter(self, debug_info, canvas):# # Visualize only the filter information# assert Falsedef visualize_mask(self, debug_info, res):return cv2.resize(debug_info[mask] * np.ones((1, 1, 3), dtypenp.float32),dsizeres,interpolationcv2.cv2.INTER_NEAREST)def draw_high_res_text(self, text, canvas):cv2.putText(canvas,text, (30, 128),cv2.FONT_HERSHEY_SIMPLEX,0.8, (0, 0, 0),thickness5)return canvasclass ExposureFilter(Filter):#gamma_param is 2*exposure_range exposure_rangedef __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name Eself.begin_filter_parameter cfg.exposure_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):return tanh_range(-self.cfg.exposure_range, self.cfg.exposure_range, initial0)(features)def process(self, img, param, defog, IcA):return img * tf.exp(param[:, None, None, :] * np.log(2))# def visualize_filter(self, debug_info, canvas):# exposure debug_info[filter_parameters][0]# if canvas.shape[0] 64:# cv2.rectangle(canvas, (8, 40), (56, 52), (1, 1, 1), cv2.FILLED)# cv2.putText(canvas, EV %.2f % exposure, (8, 48),# cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 0))# else:# self.draw_high_res_text(Exposure %.2f % exposure, canvas)class UsmFilter(Filter):#Usm_param is in [Defog_range]def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name UFself.begin_filter_parameter cfg.usm_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):return tanh_range(*self.cfg.usm_range)(features)def process(self, img, param, defog_A, IcA):def make_gaussian_2d_kernel(sigma, dtypetf.float32):radius 12x tf.cast(tf.range(-radius, radius 1), dtypedtype)k tf.exp(-0.5 * tf.square(x / sigma))k k / tf.reduce_sum(k)return tf.expand_dims(k, 1) * kkernel_i make_gaussian_2d_kernel(5)print(kernel_i.shape, kernel_i.shape)kernel_i tf.tile(kernel_i[:, :, tf.newaxis, tf.newaxis], [1, 1, 1, 1])# outputs []# for channel_idx in range(3):# data_c img[:, :, :, channel_idx:(channel_idx 1)]# data_c tf.nn.conv2d(data_c, kernel_i, [1, 1, 1, 1], SAME)# outputs.append(data_c)pad_w (25 - 1) // 2padded tf.pad(img, [[0, 0], [pad_w, pad_w], [pad_w, pad_w], [0, 0]], modeREFLECT)outputs []for channel_idx in range(3):data_c padded[:, :, :, channel_idx:(channel_idx 1)]data_c tf.nn.conv2d(data_c, kernel_i, [1, 1, 1, 1], VALID)outputs.append(data_c)output tf.concat(outputs, axis3)img_out (img - output) * param[:, None, None, :] img# img_out (img - output) * 2.5 imgreturn img_outclass UsmFilter_sigma(Filter):#Usm_param is in [Defog_range]def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name UFself.begin_filter_parameter cfg.usm_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):return tanh_range(*self.cfg.usm_range)(features)def process(self, img, param, defog_A, IcA):def make_gaussian_2d_kernel(sigma, dtypetf.float32):radius 12x tf.cast(tf.range(-radius, radius 1), dtypedtype)k tf.exp(-0.5 * tf.square(x / sigma))k k / tf.reduce_sum(k)return tf.expand_dims(k, 1) * kkernel_i make_gaussian_2d_kernel(param[:, None, None, :])print(kernel_i.shape, kernel_i.shape)kernel_i tf.tile(kernel_i[:, :, tf.newaxis, tf.newaxis], [1, 1, 1, 1])# outputs []# for channel_idx in range(3):# data_c img[:, :, :, channel_idx:(channel_idx 1)]# data_c tf.nn.conv2d(data_c, kernel_i, [1, 1, 1, 1], SAME)# outputs.append(data_c)pad_w (25 - 1) // 2padded tf.pad(img, [[0, 0], [pad_w, pad_w], [pad_w, pad_w], [0, 0]], modeREFLECT)outputs []for channel_idx in range(3):data_c padded[:, :, :, channel_idx:(channel_idx 1)]data_c tf.nn.conv2d(data_c, kernel_i, [1, 1, 1, 1], VALID)outputs.append(data_c)output tf.concat(outputs, axis3)img_out (img - output) * param[:, None, None, :] imgreturn img_outclass DefogFilter(Filter):#Defog_param is in [Defog_range]def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name DFself.begin_filter_parameter cfg.defog_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):return tanh_range(*self.cfg.defog_range)(features)def process(self, img, param, defog_A, IcA):print( defog_A:, img.shape)print( defog_A:, IcA.shape)print( defog_A:, defog_A.shape)tx 1 - param[:, None, None, :]*IcA# tx 1 - 0.5*IcAtx_1 tf.tile(tx, [1, 1, 1, 3])return (img - defog_A[:, None, None, :])/tf.maximum(tx_1, 0.01) defog_A[:, None, None, :]class GammaFilter(Filter): #gamma_param is in [-gamma_range, gamma_range]def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name Gself.begin_filter_parameter cfg.gamma_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):log_gamma_range np.log(self.cfg.gamma_range)return tf.exp(tanh_range(-log_gamma_range, log_gamma_range)(features))def process(self, img, param, defog_A, IcA):param_1 tf.tile(param, [1, 3])return tf.pow(tf.maximum(img, 0.0001), param_1[:, None, None, :])# return img# def visualize_filter(self, debug_info, canvas):# gamma debug_info[filter_parameters]# cv2.rectangle(canvas, (8, 40), (56, 52), (1, 1, 1), cv2.FILLED)# cv2.putText(canvas, G 1/%.2f % (1.0 / gamma), (8, 48),# cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 0))class ImprovedWhiteBalanceFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name Wself.channels 3self.begin_filter_parameter cfg.wb_begin_paramself.num_filter_parameters self.channelsdef filter_param_regressor(self, features):log_wb_range 0.5mask np.array(((0, 1, 1)), dtypenp.float32).reshape(1, 3)# mask np.array(((1, 0, 1)), dtypenp.float32).reshape(1, 3)print(mask.shape)assert mask.shape (1, 3)features features * maskcolor_scaling tf.exp(tanh_range(-log_wb_range, log_wb_range)(features))# There will be no division by zero here unless the WB range lower bound is 0# normalize by luminancecolor_scaling * 1.0 / (1e-5 0.27 * color_scaling[:, 0] 0.67 * color_scaling[:, 1] 0.06 * color_scaling[:, 2])[:, None]return color_scalingdef process(self, img, param, defog, IcA):return img * param[:, None, None, :]# return img# def visualize_filter(self, debug_info, canvas):# scaling debug_info[filter_parameters]# s canvas.shape[0]# cv2.rectangle(canvas, (int(s * 0.2), int(s * 0.4)), (int(s * 0.8), int(# s * 0.6)), list(map(float, scaling)), cv2.FILLED)class ColorFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.curve_steps cfg.curve_stepsself.channels int(net.shape[3])self.short_name Cself.begin_filter_parameter cfg.color_begin_paramself.num_filter_parameters self.channels * cfg.curve_stepsdef filter_param_regressor(self, features):color_curve tf.reshape(features, shape(-1, self.channels,self.cfg.curve_steps))[:, None, None, :]color_curve tanh_range(*self.cfg.color_curve_range, initial1)(color_curve)return color_curvedef process(self, img, param, defog, IcA):color_curve param# There will be no division by zero here unless the color filter range lower bound is 0color_curve_sum tf.reduce_sum(param, axis4) 1e-30total_image img * 0for i in range(self.cfg.curve_steps):total_image tf.clip_by_value(img - 1.0 * i / self.cfg.curve_steps, 0, 1.0 / self.cfg.curve_steps) * \color_curve[:, :, :, :, i]total_image * self.cfg.curve_steps / color_curve_sumreturn total_image# def visualize_filter(self, debug_info, canvas):# curve debug_info[filter_parameters]# height, width canvas.shape[:2]# for i in range(self.channels):# values np.array([0] list(curve[0][0][i]))# values / sum(values) 1e-30# scale 1# values * scale# for j in range(0, self.cfg.curve_steps):# values[j 1] values[j]# for j in range(self.cfg.curve_steps):# p1 tuple(# map(int, (width / self.cfg.curve_steps * j, height - 1 -# values[j] * height)))# p2 tuple(# map(int, (width / self.cfg.curve_steps * (j 1), height - 1 -# values[j 1] * height)))# color []# for t in range(self.channels):# color.append(1 if t i else 0)# cv2.line(canvas, p1, p2, tuple(color), thickness1)class ToneFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.curve_steps cfg.curve_stepsself.short_name Tself.begin_filter_parameter cfg.tone_begin_paramself.num_filter_parameters cfg.curve_stepsdef filter_param_regressor(self, features):tone_curve tf.reshape(features, shape(-1, 1, self.cfg.curve_steps))[:, None, None, :]tone_curve tanh_range(*self.cfg.tone_curve_range)(tone_curve)return tone_curvedef process(self, img, param, defog, IcA):# img tf.minimum(img, 1.0)# param tf.constant([[0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6], [0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6],# [0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6], [0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6],# [0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6], [0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6]])# param tf.constant([[0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6]])# param tf.reshape(# param, shape(-1, 1, self.cfg.curve_steps))[:, None, None, :]tone_curve paramtone_curve_sum tf.reduce_sum(tone_curve, axis4) 1e-30total_image img * 0for i in range(self.cfg.curve_steps):total_image tf.clip_by_value(img - 1.0 * i / self.cfg.curve_steps, 0, 1.0 / self.cfg.curve_steps) \* param[:, :, :, :, i]# p_cons [0.52, 0.53, 0.55, 1.9, 1.8, 1.7, 0.7, 0.6]# for i in range(self.cfg.curve_steps):# total_image tf.clip_by_value(img - 1.0 * i / self.cfg.curve_steps, 0, 1.0 / self.cfg.curve_steps) \# * p_cons[i]total_image * self.cfg.curve_steps / tone_curve_sumimg total_imagereturn img# def visualize_filter(self, debug_info, canvas):# curve debug_info[filter_parameters]# height, width canvas.shape[:2]# values np.array([0] list(curve[0][0][0]))# values / sum(values) 1e-30# for j in range(0, self.curve_steps):# values[j 1] values[j]# for j in range(self.curve_steps):# p1 tuple(# map(int, (width / self.curve_steps * j, height - 1 -# values[j] * height)))# p2 tuple(# map(int, (width / self.curve_steps * (j 1), height - 1 -# values[j 1] * height)))# cv2.line(canvas, p1, p2, (0, 0, 0), thickness1)class VignetFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name Vself.begin_filter_parameter cfg.vignet_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):return tf.sigmoid(features)def process(self, img, param):return img * 0 # param[:, None, None, :]def get_num_mask_parameters(self):return 5# Input: no need for tanh or sigmoid# Closer to 1 values are applied by filter more strongly# no additional TF variables insidedef get_mask(self, img, mask_parameters):with tf.name_scope(namemask):# Five parameters for one filterfilter_input_range 5assert mask_parameters.shape[1] self.get_num_mask_parameters()mask_parameters tanh_range(l-filter_input_range, rfilter_input_range,initial0)(mask_parameters)size list(map(int, img.shape[1:3]))grid np.zeros(shape[1] size [2], dtypenp.float32)shorter_edge min(size[0], size[1])for i in range(size[0]):for j in range(size[1]):grid[0, i, j,0] (i (shorter_edge - size[0]) / 2.0) / shorter_edge - 0.5grid[0, i, j,1] (j (shorter_edge - size[1]) / 2.0) / shorter_edge - 0.5grid tf.constant(grid)# (Ax)^2 (By)^2 Cinp (grid[:, :, :, 0, None] * mask_parameters[:, None, None, 0, None]) ** 2 \(grid[:, :, :, 1, None] * mask_parameters[:, None, None, 1, None]) ** 2 \mask_parameters[:, None, None, 2, None] - filter_input_range# Sharpness and inversioninp * self.cfg.maximum_sharpness * mask_parameters[:, None, None, 3,None] / filter_input_rangemask tf.sigmoid(inp)# Strengthmask * mask_parameters[:, None, None, 4,None] / filter_input_range * 0.5 0.5if not self.use_masking():print(* Masking Disabled)mask mask * 0 1else:print(* Masking Enabled)print(mask, mask.shape)return mask# def visualize_filter(self, debug_info, canvas):# brightness float(debug_info[filter_parameters][0])# cv2.rectangle(canvas, (8, 40), (56, 52), (brightness, brightness,# brightness), cv2.FILLED)#class ContrastFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name Ctself.begin_filter_parameter cfg.contrast_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):# return tf.sigmoid(features)return tf.tanh(features)def process(self, img, param, defog, IcA):luminance tf.minimum(tf.maximum(rgb2lum(img), 0.0), 1.0)contrast_lum -tf.cos(math.pi * luminance) * 0.5 0.5contrast_image img / (luminance 1e-6) * contrast_lumreturn lerp(img, contrast_image, param[:, :, None, None])# return lerp(img, contrast_image, 0.5)# def visualize_filter(self, debug_info, canvas):# exposure debug_info[filter_parameters][0]# cv2.rectangle(canvas, (8, 40), (56, 52), (1, 1, 1), cv2.FILLED)# cv2.putText(canvas, Ct %.2f % exposure, (8, 48),# cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 0))class WNBFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name BWself.begin_filter_parameter cfg.wnb_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):return tf.sigmoid(features)def process(self, img, param, defog, IcA):luminance rgb2lum(img)return lerp(img, luminance, param[:, :, None, None])# def visualize_filter(self, debug_info, canvas):# exposure debug_info[filter_parameters][0]# cv2.rectangle(canvas, (8, 40), (56, 52), (1, 1, 1), cv2.FILLED)# cv2.putText(canvas, BW%.2f % exposure, (8, 48),# cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 0))class LevelFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name Leself.begin_filter_parameter cfg.level_begin_paramself.num_filter_parameters 2def filter_param_regressor(self, features):return tf.sigmoid(features)def process(self, img, param):lower param[:, 0]upper param[:, 1] 1lower lower[:, None, None, None]upper upper[:, None, None, None]return tf.clip_by_value((img - lower) / (upper - lower 1e-6), 0.0, 1.0)# def visualize_filter(self, debug_info, canvas):# level list(map(float, debug_info[filter_parameters]))# level[1] 1# cv2.rectangle(canvas, (8, 40), (56, 52), (1, 1, 1), cv2.FILLED)# cv2.putText(canvas, %.2f %.2f % tuple(level), (8, 48),# cv2.FONT_HERSHEY_SIMPLEX, 0.25, (0, 0, 0))class SaturationPlusFilter(Filter):def __init__(self, net, cfg):Filter.__init__(self, net, cfg)self.short_name Sself.begin_filter_parameter cfg.saturation_begin_paramself.num_filter_parameters 1def filter_param_regressor(self, features):return tf.sigmoid(features)def process(self, img, param, defog, IcA):img tf.minimum(img, 1.0)hsv tf.image.rgb_to_hsv(img)s hsv[:, :, :, 1:2]v hsv[:, :, :, 2:3]# enhanced_s s (1 - s) * 0.7 * (0.5 - tf.abs(0.5 - v)) ** 2enhanced_s s (1 - s) * (0.5 - tf.abs(0.5 - v)) * 0.8hsv1 tf.concat([hsv[:, :, :, 0:1], enhanced_s, hsv[:, :, :, 2:]], axis3)full_color tf.image.hsv_to_rgb(hsv1)param param[:, :, None, None]color_param paramimg_param 1.0 - paramreturn img * img_param full_color * color_param# def visualize_filter(self, debug_info, canvas):# exposure debug_info[filter_parameters][0]# if canvas.shape[0] 64:# cv2.rectangle(canvas, (8, 40), (56, 52), (1, 1, 1), cv2.FILLED)# cv2.putText(canvas, S %.2f % exposure, (8, 48),# cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 0))# else:# self.draw_high_res_text(Saturation %.2f % exposure, canvas)util_filters.py实现
这里移除了大部分与filter.py不相干的代码仅保留lrelu, rgb2lum, tanh_range, lerp 4个函数。
def lrelu(x, leak0.2, namelrelu):with tf.variable_scope(name):f1 0.5 * (1 leak)f2 0.5 * (1 - leak)return f1 * x f2 * abs(x)def rgb2lum(image):image 0.27 * image[:, :, :, 0] 0.67 * image[:, :, :,1] 0.06 * image[:, :, :, 2]return image[:, :, :, None]def tanh01(x):return tf.tanh(x) * 0.5 0.5def tanh_range(l, r, initialNone):def get_activation(left, right, initial):def activation(x):if initial is not None:bias math.atanh(2 * (initial - left) / (right - left) - 1)else:bias 0return tanh01(x bias) * (right - left) leftreturn activationreturn get_activation(l, r, initial)def lerp(a, b, l):return (1 - l) * a l * b3、CNN-PP与DIP模块训练与优化
3.1 loss设计
这里只探讨与CNN-PP优化相关的loss。
通过观察yolov3.py中的__build_nework函数可以看到最后recovery_loss 的实现为:recovery_loss tf.reduce_sum(tf.pow(filtered_image_batch - input_data_clean, 2.0))#/(2.0 * batch_size) 其中filtered_image_batch 作为直接参与yolov3模型前向传播的对象使用filtered_image_batch 与 input_data_clean的mse loss优化CNN-PP模块的输出。故此可以看出CNN-PP | DIP实则是可以从IA-YOLO中剥离出来单独作为数据优化模块使用。
def __build_nework(self, input_data, isp_flag, input_data_clean, defog_A, IcA):filtered_image_batch input_dataself.filter_params input_datafilter_imgs_series []if isp_flag:# start_time time.time()with tf.variable_scope(extract_parameters_2):input_data tf.image.resize_images(input_data, [256, 256], methodtf.image.ResizeMethod.BILINEAR)filter_features common.extract_parameters_2(input_data, cfg, self.trainable)# filter_features tf.random_normal([1, 15], 0.5, 0.1)filters cfg.filtersfilters [x(filtered_image_batch, cfg) for x in filters]filter_parameters []for j, filter in enumerate(filters):with tf.variable_scope(filter_%d % j):print( creating filter:, j, name:, str(filter.__class__), abbr.,filter.get_short_name())print( filter_features:, filter_features.shape)filtered_image_batch, filter_parameter filter.apply(filtered_image_batch, filter_features, defog_A, IcA)filter_parameters.append(filter_parameter)filter_imgs_series.append(filtered_image_batch)print( output:, filtered_image_batch.shape)self.filter_params filter_parameters# end_time time.time()# print(filters所用时间, end_time - start_time)# input_data_shape tf.shape(input_data)# batch_size input_data_shape[0]recovery_loss tf.reduce_sum(tf.pow(filtered_image_batch - input_data_clean, 2.0))#/(2.0 * batch_size)然后模型主体返回的loss如下可见其就是基于filtered_image_batch 与input_data_clean所计算的mse loss。其中filtered_image_batch 是DIP模块基于input_data所生成的而input_data_clean不知是怎么了来的。 def compute_loss(self, label_sbbox, label_mbbox, label_lbbox, true_sbbox, true_mbbox, true_lbbox):with tf.name_scope(smaller_box_loss):loss_sbbox self.loss_layer(self.conv_sbbox, self.pred_sbbox, label_sbbox, true_sbbox,anchors self.anchors[0], stride self.strides[0])with tf.name_scope(medium_box_loss):loss_mbbox self.loss_layer(self.conv_mbbox, self.pred_mbbox, label_mbbox, true_mbbox,anchors self.anchors[1], stride self.strides[1])with tf.name_scope(bigger_box_loss):loss_lbbox self.loss_layer(self.conv_lbbox, self.pred_lbbox, label_lbbox, true_lbbox,anchors self.anchors[2], stride self.strides[2])with tf.name_scope(giou_loss):giou_loss loss_sbbox[0] loss_mbbox[0] loss_lbbox[0]with tf.name_scope(conf_loss):conf_loss loss_sbbox[1] loss_mbbox[1] loss_lbbox[1]with tf.name_scope(prob_loss):prob_loss loss_sbbox[2] loss_mbbox[2] loss_lbbox[2]with tf.name_scope(recovery_loss):recovery_loss self.recovery_lossreturn giou_loss, conf_loss, prob_loss, recovery_loss3.2 input_data_clean追溯
train相关代码
通过追溯代码train.py可以发现input_data_clean为trainset返回的图像。
此外以下代码也反应了带雾与不带雾的数据传播流程可以预见的带雾流程更加复杂。同时针对DefogFilter有各种暗通道计算方法预计是用于DefogFilter 。而据博主的了解去雾极为耗时而在IA-YOLO论文中却没有仔细说明train与train_lowlight的差异仅描述了使用CNN-PP和DIP仅增加13ms的处理时间。
而在train_lowlight.py中数据forword流程则更加简单仅使用np.power(train_data[0], lowlight_param)模拟生成低亮度数据 def train(self):self.sess.run(tf.global_variables_initializer())try:print( Restoring weights from: %s ... % self.initial_weight)self.loader.restore(self.sess, self.initial_weight)except:print( %s does not exist !!! % self.initial_weight)print( Now it starts to train YOLOV3 from scratch ...)self.first_stage_epochs 0def DarkChannel(im):b, g, r cv2.split(im)dc cv2.min(cv2.min(r, g), b);return dcdef AtmLight(im, dark):[h, w] im.shape[:2]imsz h * wnumpx int(max(math.floor(imsz / 1000), 1))darkvec dark.reshape(imsz, 1)imvec im.reshape(imsz, 3)indices darkvec.argsort(0)indices indices[(imsz - numpx):imsz]atmsum np.zeros([1, 3])for ind in range(1, numpx):atmsum atmsum imvec[indices[ind]]A atmsum / numpxreturn Adef DarkIcA(im, A):im3 np.empty(im.shape, im.dtype)for ind in range(0, 3):im3[:, :, ind] im[:, :, ind] / A[0, ind]return DarkChannel(im3)for epoch in range(1, 1self.first_stage_epochsself.second_stage_epochs):if epoch self.first_stage_epochs:train_op self.train_op_with_frozen_variableselse:train_op self.train_op_with_all_variablespbar tqdm(self.trainset)train_epoch_loss, test_epoch_loss [], []for train_data in pbar:if args.fog_FLAG:# start_time time.time()dark np.zeros((train_data[0].shape[0], train_data[0].shape[1], train_data[0].shape[2]))defog_A np.zeros((train_data[0].shape[0], train_data[0].shape[3]))IcA np.zeros((train_data[0].shape[0], train_data[0].shape[1], train_data[0].shape[2]))if DefogFilter in cfg.filters:# print(**************************)for i in range(train_data[0].shape[0]):dark_i DarkChannel(train_data[0][i])defog_A_i AtmLight(train_data[0][i], dark_i)IcA_i DarkIcA(train_data[0][i], defog_A_i)dark[i, ...] dark_idefog_A[i, ...] defog_A_iIcA[i, ...] IcA_iIcA np.expand_dims(IcA, axis-1)_, summary, train_step_loss, train_step_loss_recovery, global_step_val self.sess.run([train_op, self.write_op, self.loss, self.recovery_loss, self.global_step], feed_dict{self.input_data: train_data[0],self.defog_A: defog_A,self.IcA: IcA,self.label_sbbox: train_data[1],self.label_mbbox: train_data[2],self.label_lbbox: train_data[3],self.true_sbboxes: train_data[4],self.true_mbboxes: train_data[5],self.true_lbboxes: train_data[6],self.input_data_clean: train_data[7],self.trainable: True,})else:_, summary, train_step_loss, global_step_val self.sess.run([train_op, self.write_op, self.loss, self.global_step], feed_dict{self.input_data: train_data[7],self.label_sbbox: train_data[1],self.label_mbbox: train_data[2],self.label_lbbox: train_data[3],self.true_sbboxes: train_data[4],self.true_mbboxes: train_data[5],self.true_lbboxes: train_data[6],self.input_data_clean: train_data[7],self.trainable: True,})train_epoch_loss.append(train_step_loss)self.summary_writer.add_summary(summary, global_step_val)pbar.set_description(train loss: %.2f % train_step_loss)通过对train相关代码分析发现input_data_clean是由dataloader直接返回的
dataset相关代码
在dataset.py中的__next__函数中发现代码image, bboxes, clean_image self.parse_annotation(annotation) 再度深入追溯parse_annotation函数:
1、发现关键代码clean_image, bboxes utils.image_preporcess(np.copy(image), [self.train_input_size, self.train_input_size], np.copy(bboxes)) ,通过分析发现image_preporcess仅为图像resize函数2、通过追溯parse_annotation函数内if random.randint(0, 2) 0分支内读取图片的代码发现如下可以看到foggy_image, clean_image分别为不同的图片路径读取而来。 image cv2.imread(image_path)img_name image_path.split(/)[-1]image_name img_name.split(.)[0]image_name_index img_name.split(.)[1]bboxes np.array([list(map(lambda x: int(float(x)), box.split(,))) for box in line[1:]])if random.randint(0, 2) 0:beta random.randint(0, 9)beta 0.01 * beta 0.05if self.data_train_flag:img_name args.vocfog_traindata_dir image_name \ _ (%.2f % beta) . image_name_indexelse:img_name args.vocfog_valdata_dir image_name \ _ (%.2f % beta) . image_name_indexfoggy_image cv2.imread(img_name)clean_image image aug 代码return foggy_image, bboxes, clean_image3、通过以上分析可以发现train过程中的image实际为带雾的图像实则为data_make.py根据原始图像生成的而clean_image 则为原始图像。
4 、生成带雾图片
在IA-YOLO论文中提到了基于大气散射模型的逆向操作使用代码生成不同等级的带雾图片其代码在data_make.py将核心函数进行抽取后形成以下代码。
import numpy as np
import os
import cv2
import math
from numba import jit
import random
from PIL import Image
from ImgUilt import *#生成带雾图片i为雾气的等级
jit()
def AddHaz_loop(img_f, i):(row, col, chs) img_f.shapeA 0.5 # beta 0.08 beta 0.01 * i 0.05# 0.03size math.sqrt(max(row, col)) center (row // 2, col // 2) for j in range(row):for l in range(col):d -0.04 * math.sqrt((j - center[0]) ** 2 (l - center[1]) ** 2) sizetd math.exp(-beta * d)img_f[j][l][:] img_f[j][l][:] * td A * (1 - td)img_f np.clip(img_f*255, 0, 255).astype(np.uint8)return img_fpath rD:\实战项目\datasets\coco128\images\train2017\000000000071.jpg
image np.array(Image.open(path))
all_list[image]
for i in range(11):img_f image/255(row, col, chs) image.shapefoggy_image AddHaz_loop(img_f, i)all_list.append(foggy_image)#imgImage.fromarray(img_f)
myimshowsCL(all_list,rows4,cols3)生成的带雾图像如下所示其中myimshowsCL函数源自 python工具方法 28 2.3 单图、多图、网格显示
data_make.py
以下代码为读取voc数据生成带雾图像。
import numpy as np
import os
import cv2
import math
from numba import jit
import random# only use the image including the labeled instance objects for training
def load_annotations(annot_path):print(annot_path)with open(annot_path, r) as f:txt f.readlines()annotations [line.strip() for line in txt if len(line.strip().split()[1:]) ! 0]return annotations# print(*****************Add haze offline***************************)
def parse_annotation(annotation):line annotation.split()image_path line[0]# print(image_path)img_name image_path.split(/)[-1]# print(img_name)image_name img_name.split(.)[0]# print(image_name)image_name_index img_name.split(.)[1]# print(image_name_index)#/data/vdd/liuwenyu/data_vocfog/train/JPEGImages/if not os.path.exists(image_path):raise KeyError(%s does not exist ... %image_path)image cv2.imread(image_path)for i in range(10):jit()def AddHaz_loop(img_f, center, size, beta, A):(row, col, chs) img_f.shapefor j in range(row):for l in range(col):d -0.04 * math.sqrt((j - center[0]) ** 2 (l - center[1]) ** 2) sizetd math.exp(-beta * d)img_f[j][l][:] img_f[j][l][:] * td A * (1 - td)return img_fimg_f image/255(row, col, chs) image.shapeA 0.5 # beta 0.08 beta 0.01 * i 0.05size math.sqrt(max(row, col)) center (row // 2, col // 2) foggy_image AddHaz_loop(img_f, center, size, beta, A)img_f np.clip(foggy_image*255, 0, 255)img_f img_f.astype(np.uint8)img_name /data/vdd/liuwenyu/data_vocfog/train/JPEGImages/ image_name \ _ (%.2f%beta) . image_name_index#img_name /data/vdd/liuwenyu/data_vocfog/val/JPEGImages/ image_name \# _ (%.2f%beta) . image_name_indexcv2.imwrite(img_name, img_f)if __name__ __main__:an load_annotations(/home/liuwenyu.lwy/code/defog_yolov3/data/dataset/voc_norm_train.txt)#an load_annotations(/home/liuwenyu.lwy/code/defog_yolov3/data/dataset/voc_norm_test.txt)ll len(an)print(ll)for j in range(ll):parse_annotation(an[j])
文章转载自： http://www.morning.ttdbr.cn.gov.cn.ttdbr.cn http://www.morning.dyhlm.cn.gov.cn.dyhlm.cn http://www.morning.lnnc.cn.gov.cn.lnnc.cn http://www.morning.ctsjq.cn.gov.cn.ctsjq.cn http://www.morning.incmt.com.gov.cn.incmt.com http://www.morning.krkwp.cn.gov.cn.krkwp.cn http://www.morning.lwtfx.cn.gov.cn.lwtfx.cn http://www.morning.mznqz.cn.gov.cn.mznqz.cn http://www.morning.dkslm.cn.gov.cn.dkslm.cn http://www.morning.hrtwt.cn.gov.cn.hrtwt.cn http://www.morning.kpcky.cn.gov.cn.kpcky.cn http://www.morning.zcsch.cn.gov.cn.zcsch.cn http://www.morning.txysr.cn.gov.cn.txysr.cn http://www.morning.tstwx.cn.gov.cn.tstwx.cn http://www.morning.cprbp.cn.gov.cn.cprbp.cn http://www.morning.hrtwt.cn.gov.cn.hrtwt.cn http://www.morning.fwnyz.cn.gov.cn.fwnyz.cn http://www.morning.nfmlt.cn.gov.cn.nfmlt.cn http://www.morning.bmnm.cn.gov.cn.bmnm.cn http://www.morning.yfpnl.cn.gov.cn.yfpnl.cn http://www.morning.mrckk.cn.gov.cn.mrckk.cn http://www.morning.cfrz.cn.gov.cn.cfrz.cn http://www.morning.stpkz.cn.gov.cn.stpkz.cn http://www.morning.flfdm.cn.gov.cn.flfdm.cn http://www.morning.fmtfj.cn.gov.cn.fmtfj.cn http://www.morning.lbbrw.cn.gov.cn.lbbrw.cn http://www.morning.sfmqm.cn.gov.cn.sfmqm.cn http://www.morning.ltkms.cn.gov.cn.ltkms.cn http://www.morning.sgqw.cn.gov.cn.sgqw.cn http://www.morning.guofenmai.cn.gov.cn.guofenmai.cn http://www.morning.ljfjm.cn.gov.cn.ljfjm.cn http://www.morning.fnxzk.cn.gov.cn.fnxzk.cn http://www.morning.hrhwn.cn.gov.cn.hrhwn.cn http://www.morning.tnzwm.cn.gov.cn.tnzwm.cn http://www.morning.ljzss.cn.gov.cn.ljzss.cn http://www.morning.pgxjl.cn.gov.cn.pgxjl.cn http://www.morning.xzqzd.cn.gov.cn.xzqzd.cn http://www.morning.cwjxg.cn.gov.cn.cwjxg.cn http://www.morning.fqklt.cn.gov.cn.fqklt.cn http://www.morning.ctqlq.cn.gov.cn.ctqlq.cn http://www.morning.bzpwh.cn.gov.cn.bzpwh.cn http://www.morning.rykx.cn.gov.cn.rykx.cn http://www.morning.nxfuke.com.gov.cn.nxfuke.com http://www.morning.qhtlq.cn.gov.cn.qhtlq.cn http://www.morning.gpnwq.cn.gov.cn.gpnwq.cn http://www.morning.jltmb.cn.gov.cn.jltmb.cn http://www.morning.fgrcd.cn.gov.cn.fgrcd.cn http://www.morning.jcnmy.cn.gov.cn.jcnmy.cn http://www.morning.yxshp.cn.gov.cn.yxshp.cn http://www.morning.xdqrz.cn.gov.cn.xdqrz.cn http://www.morning.mzcsp.cn.gov.cn.mzcsp.cn http://www.morning.mwcqz.cn.gov.cn.mwcqz.cn http://www.morning.swimstaracademy.cn.gov.cn.swimstaracademy.cn http://www.morning.lpyjq.cn.gov.cn.lpyjq.cn http://www.morning.tmrjb.cn.gov.cn.tmrjb.cn http://www.morning.cgtrz.cn.gov.cn.cgtrz.cn http://www.morning.ychoise.com.gov.cn.ychoise.com http://www.morning.wqbhx.cn.gov.cn.wqbhx.cn http://www.morning.qwwhs.cn.gov.cn.qwwhs.cn http://www.morning.qxlxs.cn.gov.cn.qxlxs.cn http://www.morning.zhiheliuxue.com.gov.cn.zhiheliuxue.com http://www.morning.fbrshjf.com.gov.cn.fbrshjf.com http://www.morning.sbdqy.cn.gov.cn.sbdqy.cn http://www.morning.xxfxxf.cn.gov.cn.xxfxxf.cn http://www.morning.tbhlc.cn.gov.cn.tbhlc.cn http://www.morning.rswfj.cn.gov.cn.rswfj.cn http://www.morning.hphfy.cn.gov.cn.hphfy.cn http://www.morning.rdbj.cn.gov.cn.rdbj.cn http://www.morning.lmqfq.cn.gov.cn.lmqfq.cn http://www.morning.ndltr.cn.gov.cn.ndltr.cn http://www.morning.hmtft.cn.gov.cn.hmtft.cn http://www.morning.fllfc.cn.gov.cn.fllfc.cn http://www.morning.pffx.cn.gov.cn.pffx.cn http://www.morning.ndhxn.cn.gov.cn.ndhxn.cn http://www.morning.zxqyd.cn.gov.cn.zxqyd.cn http://www.morning.bslkt.cn.gov.cn.bslkt.cn http://www.morning.bzjpn.cn.gov.cn.bzjpn.cn http://www.morning.psdbf.cn.gov.cn.psdbf.cn http://www.morning.smfbw.cn.gov.cn.smfbw.cn http://www.morning.kpzbf.cn.gov.cn.kpzbf.cn
