def detect(im, param_vals):
scaled_ims = list(make_scaled_ims(im, model.WINDOW_SHAPE))
x, y, params = model.get_detect_model()
with tf.Session() as sess:
y_vals = []
for scaled_im in scaled_ims:
feed_dict = {x: numpy.stack([scaled_im])}
feed_dict.update(dict(zip(params, param_vals)))
y_vals.append(sess.run(y, feed_dict=feed_dict))
for i, (scaled_im, y_val) in enumerate(zip(scaled_ims, y_vals)):
for window_coords in numpy.argwhere(y_val[0, :, :, 0] >
-math.log(1./0.99 - 1)):
letter_probs = (y_val[0,
window_coords[0],
window_coords[1], 1:].reshape(
7, len(common.CHARS)))
letter_probs = common.softmax(letter_probs)
img_scale = float(im.shape[0]) / scaled_im.shape[0]
bbox_tl = window_coords * (8, 4) * img_scale
bbox_size = numpy.array(model.WINDOW_SHAPE) * img_scale
present_prob = common.sigmoid(
y_val[0, window_coords[0], window_coords[1], 0])
yield bbox_tl, bbox_tl + bbox_size, present_prob, letter_probs
def attack(im, param_vals):
scaled_ims = list(make_scaled_ims(im, model.WINDOW_SHAPE))
img = scaled_ims[2]
input = numpy.stack([img])
x_hat = tf.Variable(tf.zeros(input.shape), dtype=tf.float32)
assign_op = tf.assign(x_hat, input)
_, y, params = model.get_detect_model(x_hat)
y_val = tf.reduce_mean(y)
optim_step = tf.train.GradientDescentOptimizer(1e-1).minimize(
y_val, var_list=[x_hat])
with tf.Session(config=tf.ConfigProto()) as sess:
sess.run(assign_op)
feed_dict = {}
feed_dict.update(dict(zip(params, param_vals)))
for i in range(1):
sess.run(optim_step, feed_dict=feed_dict)
print(sess.run(y_val, feed_dict=feed_dict))
adv = x_hat.eval()
adv = adv[0, :, :]
print(adv)
print(adv.shape)
print(img)
print(img.shape)
#valid_imshow_data(adv.shape)
plt.imshow(adv)
plt.show()
def predict(list_image_path, param_values):
# Load the model which detects number plates over a sliding window.
x, y, params = model.get_detect_model()
# Execute the model at each scale.
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
for image_name in list_image_path:
im_gray = cv2.imread(image_name, cv2.IMREAD_GRAYSCALE) / 255.
im_gray = cv2.resize(im_gray, (128, 64))
print("-------------")
feed_dict = {x: numpy.stack([im_gray])}
feed_dict.update(dict(zip(params, param_values)))
y_val = sess.run(y, feed_dict=feed_dict)
letter_probs = (y_val[0,
0,
0, 1:].reshape(
10, len(common.CHARS)))
letter_probs = common.softmax(letter_probs)
present_prob = common.sigmoid(y_val[0, 0, 0, 0])
print("input", image_name)
print("output", letter_probs_to_code(letter_probs))
def detect(im, param_vals):
"""
Detect all bounding boxes of number plates in an image.
:param im:
Image to detect number plates in.
:param param_vals:
Model parameters to use. These are the parameters output by the `train`
module.
:returns:
Iterable of `bbox_tl, bbox_br, letter_probs`, defining the bounding box
top-left and bottom-right corners respectively, and a 7,36 matrix
giving the probability distributions of each letter.
"""
# Convert the image to various scales.
scaled_ims = list(make_scaled_ims(im, model.WINDOW_SHAPE))
# Load the model which detects number plates over a sliding window.
x, y, params = model.get_detect_model()
# Execute the model at each scale.
with tf.Session(config=tf.ConfigProto()) as sess:
y_vals = []
for scaled_im in scaled_ims:
feed_dict = {x: numpy.stack([scaled_im])}
feed_dict.update(dict(zip(params, param_vals)))
y_vals.append(sess.run(y, feed_dict=feed_dict))
plt.imshow(scaled_im)
plt.show()
writer = tf.summary.FileWriter("logs/", sess.graph)
# Interpret the results in terms of bounding boxes in the input image.
# Do this by identifying windows (at all scales) where the model predicts a
# number plate has a greater than 50% probability of appearing.
#
# To obtain pixel coordinates, the window coordinates are scaled according
# to the stride size, and pixel coordinates.
count_detect = 0
for i, (scaled_im, y_val) in enumerate(zip(scaled_ims, y_vals)):
for window_coords in numpy.argwhere(
y_val[0, :, :, 0] > -math.log(1. / 0.99 - 1)):
letter_probs = (y_val[0, window_coords[0], window_coords[1],
1:].reshape(7, len(common.CHARS)))
letter_probs = common.softmax(letter_probs)
img_scale = float(im.shape[0]) / scaled_im.shape[0]
bbox_tl = window_coords * (8, 4) * img_scale
bbox_size = numpy.array(model.WINDOW_SHAPE) * img_scale
present_prob = common.sigmoid(y_val[0, window_coords[0],
window_coords[1], 0])
count_detect += 1
yield bbox_tl, bbox_tl + bbox_size, present_prob, letter_probs
print("count detect:", count_detect)
print("show return window: ", bbox_tl, "return windows box: ",
bbox_tl + bbox_size)
print("present: ", present_prob)
print("letter: ", letter_probs_to_code(letter_probs))
def __init__(self, param_file):
f = numpy.load(param_file)
self.param_vals = [
f[n] for n in sorted(f.files, key=lambda s: int(s[4:]))
]
self.x, self.y, self.params = model.get_detect_model()
self.sess = tf.Session(config=tf.ConfigProto())
def detect(im, param_vals):
"""
Detect number plates in an image.
:param im:
Image to detect number plates in.
:param param_vals:
Model parameters to use. These are the parameters output by the `train`
module.
:returns:
Iterable of `bbox_tl, bbox_br, letter_probs`, defining the bounding box
top-left and bottom-right corners respectively, and a 7,36 matrix
giving the probability distributions of each letter.
"""
x, y, params = model.get_detect_model()
# Execute the model at each scale.
with tf.Session(config=tf.ConfigProto()) as sess:
feed_dict = {x: [im]}
feed_dict.update(dict(zip(params, param_vals)))
predict_values = sess.run(y, feed_dict=feed_dict)
letter_probs = predict_values.reshape(common.LENGTH, len(common.DIGITS))
letter_probs = common.softmax(letter_probs)
code= letter_probs_to_code(letter_probs)
return code
def detect(im, param_vals):
"""
Detect number plates in an image.
:param im:
Image to detect number plates in.
:param param_vals:
Model parameters to use. These are the parameters output by the `train`
module.
:returns:
Iterable of `bbox_tl, bbox_br, letter_probs`, defining the bounding box
top-left and bottom-right corners respectively, and a 7,36 matrix
giving the probability distributions of each letter.
"""
# Convert the image to various scales.
scaled_ims = list(make_scaled_ims(im, model.WINDOW_SHAPE))
# Load the model which detects number plates over a sliding window.
x, y, params = model.get_detect_model()
# Execute the model at each scale.
with tf.Session(config=tf.ConfigProto()) as sess:
y_vals = []
for scaled_im in scaled_ims:
feed_dict = {x: numpy.stack([scaled_im])}
feed_dict.update(dict(zip(params, param_vals)))
y_vals.append(sess.run(y, feed_dict=feed_dict))
# Interpret the results in terms of bounding boxes in the input image.
# Do this by identifying windows (at all scales) where the model predicts a
# number plate has a greater than 50% probability of appearing.
#
# To obtain pixel coordinates, the window coordinates are scaled according
# to the stride size, and pixel coordinates.
for i, (scaled_im, y_val) in enumerate(zip(scaled_ims, y_vals)):
for window_coords in numpy.argwhere(y_val[0, :, :, 0] >
-math.log(1./0.99 - 1)):
letter_probs = (y_val[0,
window_coords[0],
window_coords[1], 1:].reshape(
7, len(common.CHARS)))
letter_probs = common.softmax(letter_probs)
img_scale = float(im.shape[0]) / scaled_im.shape[0]
bbox_tl = window_coords * (8, 4) * img_scale
bbox_size = numpy.array(model.WINDOW_SHAPE) * img_scale
present_prob = common.sigmoid(
y_val[0, window_coords[0], window_coords[1], 0])
yield bbox_tl, bbox_tl + bbox_size, present_prob, letter_probs
def detect_presense(im, param_vals):
scaled_ims = list(make_scaled_ims(im, model.WINDOW_SHAPE))
img = scaled_ims[2]
x, y, params = model.get_detect_model()
with tf.Session(config=tf.ConfigProto()) as sess:
feed_dict = {x: numpy.stack([img])}
feed_dict.update(dict(zip(params, param_vals)))
y_val = sess.run(y, feed_dict=feed_dict)
y_val[0, :, :, 0]
if (len(numpy.argwhere(y_val[0, :, :, 0] > -math.log(1. / 0.99 - 1)))):
print('[FOUND PLATE]')
def perturb(img, param_vals):
x_origin = tf.Variable(tf.zeros(img.shape))
assign_origin_op = tf.assign(x_origin, img)
#convert image to gray so it can pass through detect model
im_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) / 255.
#resize colored image to match best scaled image shape
max_im_shape = detect_max_shape(im_gray, param_vals)
im_color_resize = tf.squeeze(
tf.image.resize_images(x_origin,
(max_im_shape[0], max_im_shape[1])))
img_gray = tf.image.rgb_to_grayscale(im_color_resize) #(250,250,1)
im_gray = tf.squeeze(tf.cast(img_gray, numpy.float32)) #(250,250)
input = tf.stack([im_gray]) #(1,250,250)
_, y, params = model.get_detect_model(input)
#mean over reduced probability for presence and letter detection, y[:,:,:,0] for just presence
y_mean = tf.reduce_mean(y)
optim_step = tf.train.GradientDescentOptimizer(3).minimize(
y_mean, var_list=[x_origin])
adv = []
#create bounds for how far the image can deviate
epsilon = tf.placeholder(tf.float32, ())
below = img - epsilon
above = img + epsilon
projected = tf.clip_by_value(tf.clip_by_value(x_origin, below, above),
0, 255)
with tf.control_dependencies([projected]):
project_step = tf.assign(x_origin, projected)
#training
with tf.Session(config=tf.ConfigProto()) as sess:
sess.run(assign_origin_op)
feed_dict = {}
feed_dict.update(dict(zip(params, param_vals)))
for i in range(400):
sess.run(optim_step, feed_dict=feed_dict)
sess.run(project_step, feed_dict={epsilon: 8000 / 255.0})
print(sess.run(y_mean, feed_dict=feed_dict))
if (isDetected(x_origin.eval().astype(numpy.uint8),
param_vals) == False):
break
adv = (x_origin.eval().astype(numpy.uint8))
return adv
def isDetected(im, param_vals):
im_gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY) / 255.
scaled_ims = list(make_scaled_ims(im_gray, model.WINDOW_SHAPE))
x, y, params = model.get_detect_model()
with tf.Session(config=tf.ConfigProto()) as sess:
y_vals = []
for scaled_im in scaled_ims:
feed_dict = {x: numpy.stack([scaled_im])}
feed_dict.update(dict(zip(params, param_vals)))
y_vals.append(sess.run(y, feed_dict=feed_dict))
count = 0
for i, (scaled_im, y_val) in enumerate(zip(scaled_ims, y_vals)):
for window_coords in numpy.argwhere(
y_val[0, :, :, 0] > -math.log(1. / 0.99 - 1)):
count = count + 1 #only care about number of boxes it can detect
print("COUNT: ", count)
return count != 0
def detect(im, param_vals):
# Convert the image to various scales.
scaled_ims = list(make_scaled_ims(im, model.WINDOW_SHAPE))
# Load the model which detects number plates over a sliding window.
x, y, params = model.get_detect_model()
# Execute the model at each scale.
with tf.Session(config=tf.ConfigProto()) as sess:
y_vals = []
for scaled_im in scaled_ims:
feed_dict = {x: numpy.stack([scaled_im])}
feed_dict.update(dict(zip(params, param_vals)))
y_vals.append(sess.run(y, feed_dict=feed_dict))
# Interpret the results in terms of bounding boxes in the input image.
# Do this by identifying windows (at all scales) where the model predicts a
# number plate has a greater than 50% probability of appearing.
#
# To obtain pixel coordinates, the window coordinates are scaled according
# to the stride size, and pixel coordinates.
for i, (scaled_im, y_val) in enumerate(zip(scaled_ims, y_vals)):
#print(i)
#print(numpy.argwhere(y_val[0, :, :, 0] > -math.log(1./0.99 - 1)))
#print(-math.log(1./0.99 - 1))
#print(numpy.argwhere(y_val[0, :, :, 0] >-math.log(1./0.99 - 1)))
for window_coords in numpy.argwhere(
y_val[0, :, :, 0] > -math.log(1. / 0.99 - 1)):
letter_probs = (y_val[0, window_coords[0], window_coords[1],
1:].reshape(7, len(common.CHARS)))
letter_probs = common.softmax(letter_probs)
img_scale = float(im.shape[0]) / scaled_im.shape[0]
bbox_tl = window_coords * (8, 4) * img_scale
bbox_size = numpy.array(model.WINDOW_SHAPE) * img_scale
present_prob = common.sigmoid(y_val[0, window_coords[0],
window_coords[1], 0])
yield bbox_tl, bbox_tl + bbox_size, present_prob, letter_probs
def detect(im, param_vals):
"""
Detect number plates in an image.
:param im:
Image to detect number plates in.
:param param_vals:
Model parameters to use. These are the parameters output by the `train`
module.
:returns:
Iterable of `bbox_tl, bbox_br, letter_probs`, defining the bounding box
top-left and bottom-right corners respectively, and a 7,36 matrix
giving the probability distributions of each letter.
"""
# Convert the image to various scales.
scaled_ims = list(make_scaled_ims(im, model.WINDOW_SHAPE))
for i, im in enumerate(scaled_ims):
cv2.imwrite("scale{}.bmp".format(i), im*255.0)
# Load the model which detects number plates over a sliding window.
x, y, params = model.get_detect_model()
# Execute the model at each scale.
with tf.Session(config=tf.ConfigProto()) as sess:
y_vals = []
for scaled_im in scaled_ims:
feed_dict = {x: numpy.stack([scaled_im])}
feed_dict.update(dict(zip(params, param_vals)))
y_vals.append(sess.run(y, feed_dict=feed_dict))
# Interpret the results in terms of bounding boxes in the input image.
# Do this by identifying windows (at all scales) where the model predicts a
# number plate has a greater than 50% probability of appearing.
#
# To obtain pixel coordinates, the window coordinates are scaled according
# to the stride size, and pixel coordinates.
for i, (scaled_im, y_val) in enumerate(zip(scaled_ims, y_vals)):
#for window_coords in numpy.argwhere(y_val[0, :, :, 0] >
# -math.log(1./0.99 - 1)):
# win = numpy.argmax(y_val[0, :, :, 0])
# print "win point:{} {} max:{}".format(win/y_val.shape[1], win%y_val.shape[1],
# y_val[0, win/y_val.shape[1], win%y_val.shape[1], 0])
for window_coords in numpy.argwhere(y_val[0, :, :, 0] > 0):
p_probs = y_val[0, window_coords[0], window_coords[1], 0]
# print "p_probs:{}".format(p_probs)
letter_probs = (y_val[0,
window_coords[0],
window_coords[1], 1:].reshape(
7, len(common.CHARS)))
letter_probs = common.softmax(letter_probs)
img_scale = float(im.shape[0]) / scaled_im.shape[0]
bbox_tl = window_coords * (16, 8) * img_scale
bbox_size = numpy.array(model.WINDOW_SHAPE) * img_scale
present_prob = common.sigmoid(
y_val[0, window_coords[0], window_coords[1], 0])
yield bbox_tl, bbox_tl + bbox_size, present_prob, letter_probs
def detect(im, param_vals):
"""
Detect number plates in an image.
:param im:
Image to detect number plates in.
:param param_vals:
Model parameters to use. These are the parameters output by the `train`
module.
:returns:
Iterable of `bbox_tl, bbox_br, letter_probs`, defining the bounding box
top-left and bottom-right corners respectively, and a 7,36 matrix
giving the probability distributions of each letter.
"""
# Convert the image to various scales.
scaled_ims = list(make_scaled_ims(im,
model.WINDOW_SHAPE)) # 针对需检测图片产生迭代缩放图片列表
# Load the model which detects number plates over a sliding window.
x, y, params = model.get_detect_model() # 获取检测网络(5个卷积层)的输入,输出以及所有参数
# Execute the model at each scale.
with tf.Session(config=tf.ConfigProto()) as sess: # 建立会话
y_vals = []
for scaled_im in scaled_ims: # 遍历缩放图片列表
feed_dict = {x: numpy.stack([scaled_im])} # 建立字典feed_dict{输入：缩放图片}
feed_dict.update(dict(zip(
params,
param_vals))) # 将字典{参数：参数值}更新到字典feed_dict中->{输入：缩放图片, 参数：参数值}
y_vals.append(sess.run(y,
feed_dict=feed_dict)) # 提供检测图片和网络参数值,执行检测操作
# Interpret the results in terms of bounding boxes in the input image.
# Do this by identifying windows (at all scales) where the model predicts a
# number plate has a greater than 50% probability of appearing.
#
# To obtain pixel coordinates, the window coordinates are scaled according
# to the stride size, and pixel coordinates.
for i, (scaled_im,
y_val) in enumerate(zip(scaled_ims,
y_vals)): # 枚举元组(缩放图片, 检测输出)及对应下标
for window_coords in numpy.argwhere(
y_val[0, :, :, 0] > -math.log(1. / 0.99 - 1)
): # 对应sigmoid输出号码牌存在概率大于0.99的下标
letter_probs = (y_val[0, window_coords[0], window_coords[1],
1:].reshape(7, len(common.CHARS)))
letter_probs = common.softmax(letter_probs) # 车牌号识别正确的概率分布
img_scale = float(im.shape[0]) / scaled_im.shape[0] # 缩放比例
bbox_tl = window_coords * (
8, 4) * img_scale # 计算缩放图片上号码牌的boudingbox的左上角坐标
bbox_size = numpy.array(
model.WINDOW_SHAPE) * img_scale # 号码牌的boudingbox的尺寸
present_prob = common.sigmoid(y_val[0, window_coords[0],
window_coords[1],
0]) # 号码牌是否存在的概率分布
yield bbox_tl, bbox_tl + bbox_size, present_prob, letter_probs # 返回号码牌boudingbox的左上角坐标,右下角坐标,
)
import collections
import itertools
import math
import sys
import cv2
import numpy
import tensorflow as tf
import common
import model
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.50)
sess1 = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
x, y, params = model.get_detect_model()
f = numpy.load("weights.npz")
param_vals = [f[n] for n in sorted(f.files, key=lambda s: int(s[4:]))]
def make_scaled_ims(im, min_shape):
ratio = 1. / 2**0.5
shape = (im.shape[0] / ratio, im.shape[1] / ratio)
#(479.4183976444793, 388.90872965260115)#(im.shape[0] / ratio, im.shape[1] / ratio)
print("MIN SHAPE", shape, min_shape)
while True:
shape = (int(shape[0] * ratio), int(shape[1] * ratio))
if shape[0] < min_shape[0] or shape[1] < min_shape[1]:
break