def yolo3_data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes, enhance_augment, rescale_interval, multi_anchor_assign):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
# prepare multiscale config
rescale_step = 0
input_shape_list = get_multiscale_list()
while True:
if rescale_interval > 0:
# Do multi-scale training on different input shape
rescale_step = (rescale_step + 1) % rescale_interval
if rescale_step == 0:
input_shape = input_shape_list[random.randint(0, len(input_shape_list)-1)]
image_data = []
box_data = []
for b in range(batch_size):
if i==0:
np.random.shuffle(annotation_lines)
image, box = get_ground_truth_data(annotation_lines[i], input_shape, augment=True)
image_data.append(image)
box_data.append(box)
i = (i+1) % n
image_data = np.array(image_data)
box_data = np.array(box_data)
if enhance_augment == 'mosaic':
# add random mosaic augment on batch ground truth data
image_data, box_data = random_mosaic_augment(image_data, box_data, prob=0.2)
y_true = preprocess_true_boxes(box_data, input_shape, anchors, num_classes, multi_anchor_assign)
yield [image_data, *y_true], np.zeros(batch_size)
def yolo2_data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes, rescale_interval):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
# prepare multiscale config
rescale_step = 0
input_shape_list = get_multiscale_list()
while True:
if rescale_interval > 0:
# Do multi-scale training on different input shape
rescale_step = (rescale_step + 1) % rescale_interval
if rescale_step == 0:
input_shape = input_shape_list[random.randint(0,len(input_shape_list)-1)]
image_data = []
box_data = []
for b in range(batch_size):
if i==0:
np.random.shuffle(annotation_lines)
image, box = get_ground_truth_data(annotation_lines[i], input_shape, augment=True)
image_data.append(image)
box_data.append(box)
i = (i+1) % n
image_data = np.array(image_data)
box_data = np.array(box_data)
y_true_data = get_y_true_data(box_data, anchors, input_shape, num_classes)
yield [image_data, y_true_data], np.zeros(batch_size)
def __init__(self,
annotation_lines,
batch_size,
input_shape,
anchors,
num_classes,
enhance_augment=None,
rescale_interval=-1,
shuffle=True):
self.annotation_lines = annotation_lines
self.batch_size = batch_size
self.input_shape = input_shape
self.anchors = anchors
self.num_classes = num_classes
self.enhance_augment = enhance_augment
self.indexes = np.arange(len(self.annotation_lines))
self.shuffle = shuffle
# prepare multiscale config
# TODO: error happens when using Sequence data generator with
# multiscale input shape, disable multiscale first
if rescale_interval != -1:
raise ValueError(
"tf.keras.Sequence generator doesn't support multiscale input, pls remove related config"
)
#self.rescale_interval = rescale_interval
self.rescale_interval = -1
self.rescale_step = 0
self.input_shape_list = get_multiscale_list()
def __init__(self, annotation_lines, batch_size, input_shape, anchors, num_classes, rescale_interval=-1, shuffle=True):
self.annotation_lines = annotation_lines
self.batch_size = batch_size
self.input_shape = input_shape
self.anchors = anchors
self.num_classes = num_classes
self.indexes = np.arange(len(self.annotation_lines))
self.shuffle = shuffle
# prepare multiscale config
# TODO: error happens when using Sequence data generator with
# multiscale input shape, disable multiscale first
#self.rescale_interval = rescale_interval
self.rescale_interval = -1
self.rescale_step = 0
self.input_shape_list = get_multiscale_list()
