def predcit(self, img, label, index):
if self.debug:
cv2.circle(img[0], (int(label[0]), int(label[1])), 3, (255, 0, 0),
-1)
print "img.shape: ", img.shape
print "label.shape: ", label.shape
pred = self.model.predict(img, verbose=1)
print "pred.shape: ", pred.shape
img = ut.plotTarget(img[0], label, ifSquareOnly=True, ifGreen=True)
# img = ut.plotTarget(img, ut.deNormalize(labels, self.imSize), self.imSize, ifSquareOnly = True, ifGreen = True)
# cv2.imwrite(self.evaluationOutputDir + 'inputTestImg' + str(saveCount) + '.jpg', img)
labelImg = ut.plotTarget(img,
ut.deNormalize(pred[0], self.imSize),
self.imSize,
ifSquareOnly=True)
# cv2.imshow("img", img)
# cv2.waitKey(0)
print "saving" + self.evaluationOutputDir + str(
index) + 'Pred' + '.jpg'
# cv2.imwrite(self.evaluationOutputDir + str(index) + 'Input' + '.jpg', img)
cv2.imwrite(self.evaluationOutputDir + str(index) + 'Pred' + '.jpg',
labelImg)
def predcit(self):
saveCount = 0
for iter in range(self.MaxTestIters):
test_start = iter * self.batch_size
test_end = (iter + 1) * self.batch_size
# if iter == self.MaxTestIters - 1:
# test_end = len(self.DataTe)
print "test_start, test_end: ", test_start, test_end
if self.ifMenpo39DataSet:
X_batch_T, label_BB_T, Z_Names_T = self.DataGenBB(
train_start=test_start, train_end=test_end)
elif self.ifpreProcessedSemifrontal:
X_batch_T, label_BB_T, Z_Names_T = self.DataGenBB(
train_start=test_start, train_end=test_end)
else:
X_batch_T, label_BB_T, Z_Names_T = self.DataGenBB(
DataStrs=self.DataTe,
train_start=test_start,
train_end=test_end)
# print "X_batch_T.shape: ", X_batch_T.shape
# print "label_BB_T.shape: ", label_BB_T.shape
pred = self.model.predict(X_batch_T, verbose=1)
# print "type(pred): ", type(pred)
# print "pred.shape: ", pred.shape
# print "Z_Names_T.shape: ", Z_Names_T.shape
# print "X_batch_T[-1]: ", X_batch_T[-1]
# print "label_BB_T[-1]: ", label_BB_T[-1]
# raise "debug"
for i in range(self.batch_size):
labels = label_BB_T[i]
img = X_batch_T[i]
index = Z_Names_T[i].split(" ")[0]
img = ut.plotTarget(img,
labels,
self.imSize,
ifSquareOnly=True,
ifGreen=True)
# img = ut.plotTarget(img, ut.deNormalize(labels, self.imSize), self.imSize, ifSquareOnly = True, ifGreen = True)
# cv2.imwrite(self.evaluationOutputDir + 'inputTestImg' + str(saveCount) + '.jpg', img)
# labelImg = ut.plotTarget(img, ut.deNormalize(pred[i], self.imSize), self.imSize, ifSquareOnly = True)
cv2.imshow("img", img)
cv2.waitKey(0)
# print 'save predTestLabelImg' + str(saveCount) + '.jpg to: ' + self.evaluationOutputDir
cv2.imwrite(
self.evaluationOutputDir + str(index) + 'Input' + '.jpg',
img)
# cv2.imwrite(self.evaluationOutputDir + str(index) + 'Pred' + '.jpg', labelImg)
saveCount += 1
def loop(self):
for i in range(len(self.imgs)):
imgName = self.imgs[i]
# print "imgName: ", imgName
fileHeader = imgName.split('.')[0]
print "self.ImgDir + imgName: ", self.ImgDir + imgName
img = cv2.imread(self.ImgDir + imgName)
print "self.labelDir + imgNameHeader + .txt: ", self.labelDir + fileHeader + ".txt"
label = np.loadtxt(self.labelDir + fileHeader + ".txt")
if self.debug:
img = ut.plotTarget(img,
label,
ifSquareOnly=True,
ifGreen=True)
cv2.imshow("img", img)
cv2.waitKey(0)
if img != None:
# print "img.shape: ", img.shape
w, h, _ = img.shape
xMean, yMean, edge = label
xMean = xMean / w
yMean = yMean / h
edge = edge / w
img = cv2.resize(img, (self.imSize, self.imSize))
xMean = xMean * self.imSize
yMean = yMean * self.imSize
edge = edge * self.imSize
label = np.asarray([xMean, yMean, edge])
else:
print "cannot find : ", imgName
InputData = np.zeros([1, self.imSize, self.imSize, 3],
dtype=np.float32)
InputLabel = np.zeros([1, 3], dtype=np.float32)
InputData[0] = img
self.predcit(InputData, label, fileHeader)
img = Image.fromarray(img.astype(np.uint8))
cropImg = img.crop((newXMin, newYMin, newXMax, newYMax))
cropImg = np.array(cropImg)
if debug:
cv2.circle(cropImg, (int(xMean), int(yMean)), 2, (0, 0, 255), -1)
label = np.asarray([xMean, yMean, edge])
# # NOTE: its img[y: y + h, x: x + w] and *not* img[x: x + w, y: y + h]
if debug:
img = plotTarget(img, [newXMin, newYMin, newXMax, newYMax])
img = ut.plotTarget(img, [xMean, yMean, newEdge],
ifSquareOnly=True)
# img = ut.plotTarget(img, label, ifSquareOnly = True, ifGreen = True)
# img = plotTarget(cropImg, [xMean - xEdge/2.0, yMean - yEdge/2.0, xMean + xEdge/2.0, yMean + yEdge/2.0], ifGreen = True)
img = ut.plotTarget(cropImg,
label,
ifSquareOnly=True,
ifGreen=True)
img = ut.plotTarget(img, label, ifSquareOnly=True, ifGreen=True)
cv2.imshow("img", img)
cv2.imshow("cropped", cropImg)
cv2.waitKey(0)
np.savetxt(labelOutputDir + fileHeader + '.txt', label)
cv2.imwrite(imgOutputDir + fileHeader + '.jpg', cropImg)
import os
import cv2
import numpy as np
import random
import utility as ut
files = ["2430.rec"]
ImgDir = 'data/preProcessedSemifrontal/img/'
labelDir = 'data/preProcessedSemifrontal/label/'
imgName = "2430.jpg"
fileHeader = imgName.split('.')[0]
print "self.ImgDir + imgName: ", ImgDir + imgName
img = cv2.imread(ImgDir + imgName)
print "self.labelDir + imgNameHeader + .txt: ", labelDir + fileHeader + ".txt"
label = np.loadtxt(labelDir + fileHeader + ".txt")
img = ut.plotTarget(img, label, ifSquareOnly=True, ifGreen=True)
cv2.imshow("img", img)
cv2.waitKey(0)
print "pts: ", pts
print "w, h: ", w, h
x1, y1, x2, y2 = [int(i) for i in pts]
x1, x2 = x1 / w, x2 / w
y1, y2 = y1 / h, y2 / h
w, h, _ = img.shape
x1, x2 = x1 * w, x2 * w
y1, y2 = y1 * h, y2 * h
x_mean = (x1 + x2) / float(2)
y_mean = (y1 + y2) / float(2)
edge = max(y2 - y1, x2 - x1) * 1.5
new_img = ut.plotTarget(img, (x_mean, y_mean, edge),
ifSquareOnly=True)
print "int(y_mean - edge/2.0), int(y_mean + edge/2.0), int(x_mean - edge/2.0), int(x_mean + edge/2.0)"
print int(y_mean - edge / 2.0), int(y_mean + edge / 2.0), int(
x_mean - edge / 2.0), int(x_mean + edge / 2.0)
y_min = int(y_mean -
edge / 2.0) if int(y_mean - edge / 2.0) >= 0 else 0
y_max = int(y_mean +
edge / 2.0) if int(y_mean + edge / 2.0) <= w else w
x_min = int(x_mean -
edge / 2.0) if int(x_mean - edge / 2.0) >= 0 else 0
x_max = int(x_mean +
edge / 2.0) if int(x_mean + edge / 2.0) <= h else h
print "y_min, y_max, x_min, x_max: ", y_min, y_max, x_min, x_max
def train_on_batch(self, nb_epoch, MaxIters):
if os.path.exists(self.modelDir) == False:
os.mkdir(self.modelDir)
testCount = 0
trainCount = 0
for e in range(nb_epoch):
shuffle(self.DataTr)
iterTest = 0
for iter in range(self.MaxIters):
train_start = iter * self.batch_size
train_end = (iter + 1) * self.batch_size
X_batch, label_BB, Z_Names = self.DataGenBB(
self.DataTr, train_start=train_start, train_end=train_end)
# for i in range(self.batch_size):
# labels = label_BB[i]
# img = X_batch[i]
# labelImg = ut.plotTarget(img, ut.deNormalize(labels, self.imSize), self.imSize, ifSquareOnly = True)
# cv2.imwrite(self.outputDir + 'inputTrainlabelImg' + str(trainCount) + '.jpg', labelImg)
loss, tras, pred = self.model.train_on_batch(X_batch, label_BB)
trainCount += 1
if trainCount >= 20:
trainCount = 0
# if loss in [None, float("inf"), float("-inf"), Decimal('Infinity')] or "nan" in str(loss):
# print "-----model reset weights-----"
# self.reset_weights(self.model)
# self.reset_model(self.model)
# self.model.reset_states()
# print "*****"
print "loss, train: ", loss
if iter % 1000 == 0:
logInfo = ""
if os.path.exists(self.outputDir +
'log.txt') and self.init == False:
f = open(self.outputDir + 'log.txt', 'a')
else:
f = open(self.outputDir + 'log.txt', 'w')
self.init = False
iterationInfo = ("^^^^^" + "\n" + 'iteration: ' +
str(iter))
logInfo += iterationInfo
print iterationInfo
# labelImg = ut.plotTarget(X_batch[0], pred[0])
index = random.randint(0, self.batch_size - 1)
labelImg = ut.plotTarget(X_batch[index],
ut.deNormalize(
pred[index], self.imSize),
self.imSize,
ifSquareOnly=True)
print 'save predTrainLabelImg' + str(
testCount) + '.jpg to: ' + self.outputDir
# cv2.imwrite(self.outputDir + 'predTrainLabelImg' + str(testCount) + '.jpg', labelImg)
test_start = iterTest * self.batch_size
test_end = (iterTest + 1) * self.batch_size
X_batch_T, label_BB_T, Z_Names_T = self.DataGenBB(
self.DataTe,
train_start=test_start,
train_end=test_end)
loss, tras, pred = self.model.evaluate(
X_batch_T, label_BB_T)
testCount += 1
if testCount >= 20:
testCount = 0
index = random.randint(0, self.batch_size - 1)
labelImg = ut.plotTarget(X_batch_T[index],
ut.deNormalize(
pred[index], self.imSize),
self.imSize,
ifSquareOnly=True)
print 'save predTestLabelImg' + str(
testCount) + '.jpg to: ' + self.outputDir
# cv2.imwrite(self.outputDir + 'predTestLabelImg' + str(testCount) + '.jpg', labelImg)
testInfo = ("====" + "\n" + "loss, TEST: " + str(loss))
logInfo += testInfo
print testInfo
iterTest += self.batch_size
iterTest %= self.MaxTestIters
f.write(logInfo)
f.close()