def preprocess(self, image, gt_boxes):
image, gt_boxes = utils.resize_image_correct_bbox(image, gt_boxes, self.image_h, self.image_w)
if self.debug: return image, gt_boxes
y_true_13, y_true_26, y_true_52 = tf.py_func(self.preprocess_true_boxes, inp=[gt_boxes],
Tout=[tf.float32, tf.float32, tf.float32])
image = image / 255.
# image , cellxcellx3x(4+1+class_id)
return image, y_true_13, y_true_26, y_true_52
def preprocess(self, image, gt_boxes):
image, gt_boxes = utils.resize_image_correct_bbox(
image, gt_boxes, self.IMAGE_H, self.IMAGE_W)
if self.DEBUG: return image, gt_boxes
# data augmentation (continue to work)
image = image / 255.
y_true, true_boxes = tf.py_func(self.preprocess_true_boxes,
inp=[gt_boxes],
Tout=[tf.float32, tf.float32])
return image, y_true, true_boxes
def preprocess(self, image, gt_boxes):
image, gt_boxes = utils.resize_image_correct_bbox(
image, gt_boxes, self.image_h, self.image_w)
if self.debug:
return image, gt_boxes
y_true = tf.py_func(self.preprocess_true_boxes,
inp=[gt_boxes],
Tout=[tf.float32])
return image, y_true
def preprocess(self, image, gt_boxes):
# resize_image_correct_bbox
image, gt_boxes = utils.resize_image_correct_bbox(
image, gt_boxes, self.image_h, self.image_w)
if self.debug: return image, gt_boxes
y_true_13, y_true_26, y_true_52 = tf.py_func(
self.preprocess_true_boxes,
inp=[gt_boxes],
Tout=[tf.float32, tf.float32, tf.float32])
# data augmentation
# pass
image = image / 255.
return image, y_true_13, y_true_26, y_true_52
def preprocess(self, image, gt_boxes):
################################# data augmentation ##################################
# data_aug_flag = tf.to_int32(tf.random_uniform(shape=[], minval=-5, maxval=5))
# caseO = tf.equal(data_aug_flag, 1), lambda: self.flip_left_right(image, gt_boxes)
# case1 = tf.equal(data_aug_flag, 2), lambda: self.random_distort_color(image, gt_boxes)
# case2 = tf.equal(data_aug_flag, 3), lambda: self.random_blur(image, gt_boxes)
# case3 = tf.equal(data_aug_flag, 4), lambda: self.random_crop(image, gt_boxes)
# image, gt_boxes = tf.case([caseO, case1, case2, case3], lambda: (image, gt_boxes))
image, gt_boxes = utils.resize_image_correct_bbox(image, gt_boxes, self.image_h, self.image_w)
if self.debug: return image, gt_boxes
y_true_13, y_true_26, y_true_52 = tf.py_func(self.preprocess_true_boxes, inp=[gt_boxes],
Tout = [tf.float32, tf.float32, tf.float32])
image = image / 255.
return image, y_true_13, y_true_26, y_true_52
def preprocess(self, image, gt_boxes):
################################# data augmentation ##################################
is_flip = tf.random_uniform(shape=[],
minval=0,
maxval=2,
dtype=tf.int32)
is_distort = tf.random_uniform(shape=[],
minval=0,
maxval=2,
dtype=tf.int32)
is_blur = tf.random_uniform(shape=[],
minval=0,
maxval=2,
dtype=tf.int32)
image, gt_boxes = tf.cond(
tf.equal(is_flip,
1), lambda: self.flip_left_right(image, gt_boxes), lambda:
(image, gt_boxes))
image, gt_boxes = tf.cond(
tf.equal(is_distort,
1), lambda: self.random_distort_color(image, gt_boxes),
lambda: (image, gt_boxes))
image, gt_boxes = tf.cond(tf.equal(is_blur, 1),
lambda: self.random_blur(image, gt_boxes),
lambda: (image, gt_boxes))
image, gt_boxes = utils.resize_image_correct_bbox(
image, gt_boxes, self.image_h, self.image_w)
if self.debug: return image, gt_boxes
y_true = tf.py_func(self.preprocess_true_boxes,
inp=[gt_boxes],
Tout=tf.float32)
image = image / 255.
return image, y_true
model = yolov3.yolov3(80)
with tf.variable_scope('yolov3'):
feature_maps = model.forward(inputs, is_training=True)
load_ops = utils.load_weights(tf.global_variables(scope='yolov3'),
"./checkpoint/yolov3.weights")
sess.run(load_ops)
loss = model.compute_loss(feature_maps, [y_true_13, y_true_26, y_true_52])
optimizer = tf.train.GradientDescentOptimizer(0.001)
train_op = optimizer.minimize(loss)
sess.run(tf.global_variables_initializer())
for image_path in dataset.keys():
image = Image.open(image_path)
true_boxes, true_labels = dataset[image_path]
image, true_boxes = utils.resize_image_correct_bbox(
image, true_boxes, input_shape)
scores = np.ones(len(true_boxes))
# utils.draw_boxes(image, boxes, scores, labels, classes)
true_boxes = np.expand_dims(true_boxes, 0)
true_labels = np.expand_dims(true_labels, 0)
y_true = utils.preprocess_true_boxes(true_boxes, true_labels, input_shape,
anchors, num_classes)
image_data = np.expand_dims(np.array(image, dtype=np.float32) / 255.,
axis=0)
_, result = sess.run(
[train_op, loss],
feed_dict={
inputs: image_data,
y_true_13: y_true[0],