def data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes,teacher):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
while True:
#print("data"+str(i) )
image_data = []
box_data = []
for b in range(batch_size):
#if i==0:
# np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i], input_shape, random=False)
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 = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)
m_true = teacher.predict(image_data)
#l_true = [ np.zeros( shape=( batch_size ,416//{0:32, 1:16, 2:8}[l], 416//{0:32, 1:16, 2:8}[l], 9//3, 20+5) ) for l in range(3) ]
anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if len(m_true)==3 else [[3,4,5], [1,2,3]]
for l in range( len(m_true) ) :
pred_xy, pred_wh , pred_conf , pred_class = numpy_yolo_head( m_true[l] ,anchors[anchor_mask[l]], input_shape )
pred_detect = np.concatenate([pred_xy, pred_wh , pred_conf , pred_class ],axis=-1)
#print("inside")
box = np.where(y_true[l][...,4] > 0.5 )
box = np.transpose(box)
for k in range(len(box)):
m_true[l][tuple(box[k])] = pred_detect[tuple(box[k])]
yield image_data, y_true ,m_true
def data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes,model):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
while True:
image_data = []
box_data = []
for b in range(batch_size):
if i==0:
np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i], input_shape, random=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 = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)
# print(image_data.shape)
y_true = model.predict(image_data)
# print("d")
# print(y_true[0].shape)
# print(y_true[1].shape)
# print(y_true[2].shape)
# y_true[0] = y_true[0].reshape(y_true[0].shape[0], y_true[0].shape[1], y_true[0].shape[2], 3 , y_true[0].shape[3]//3 )
# y_true[1] = y_true[1].reshape(y_true[1].shape[0], y_true[1].shape[1], y_true[1].shape[2], 3 , y_true[1].shape[3]//3 )
# y_true[2] = y_true[2].reshape(y_true[2].shape[0], y_true[2].shape[1], y_true[2].shape[2], 3 , y_true[2].shape[3]//3 )
yield [image_data, *y_true]#, np.zeros(batch_size)
def data_generator(annotation_lines, batch_size, input_shape, anchors,
num_classes, teacher):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
while True:
image_data = []
box_data = []
for b in range(batch_size):
if i == 0:
np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i],
input_shape,
random=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 = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)
box = teacher.predict(image_data)
#np.savez('temp_logits.npz', 0 = box[0] , 1=box[1] , 2=box[2] )
#box = np.load('temp_logits.npz')[()]
#y_true = box
y_true = box
yield [image_data, *y_true], np.zeros(batch_size)
def data_generator(annotation_line, batch_size, input_shape, anchors,
num_classes):
"""
训练数据生成器
:param annotation_line: 图片名与标注
:param batch_size: 批大小
:param input_shape: 输入尺寸
:param anchors: 锚框坐标
:param num_classes: 类别数
:return: 生成器，输出每个batch的image data和y_true
"""
n = len(annotation_line)
if n == 0 or batch_size <= 0:
return None
else:
i = 0
while True:
image_data = []
box_data = []
for batch in range(batch_size):
if i == 0:
np.random.shuffle(annotation_line) # 对图片进行乱序
# 对每张图片进行数据增强，并分开图片地址和标签
image, box = get_random_data(annotation_line[i],
input_shape,
random=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 = preprocess_true_boxes(box_data, input_shape, anchors,
num_classes)
yield [image_data, *y_true], np.zeros(batch_size)
def data_generator(annotation_lines, batch_size, input_shape, anchors,
num_classes, teacher):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
while True:
image_data = []
box_data = []
for b in range(batch_size):
if i == 0:
np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i],
input_shape,
random=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 = preprocess_true_boxes(box_data, input_shape, anchors,
num_classes)
m_true = teacher.predict(image_data)
anchor_mask = [[6, 7, 8], [3, 4, 5], [0, 1, 2]
] if len(m_true) == 3 else [[3, 4, 5], [1, 2, 3]]
for l in range(len(m_true)):
anchors_tensor = np.reshape(
anchors[anchor_mask[l]],
[1, 1, 1, len(anchors[anchor_mask[l]]), 2])
grid_shape = m_true[l].shape[1:3] # height, width
grid_shape
grid_y = np.tile(
np.reshape(np.arange(0, stop=grid_shape[0]), [-1, 1, 1, 1]),
[1, grid_shape[1], 1, 1])
grid_x = np.tile(
np.reshape(np.arange(0, stop=grid_shape[1]), [1, -1, 1, 1]),
[grid_shape[0], 1, 1, 1])
grid = np.concatenate([grid_x, grid_y], axis=-1)
#print(l)
m_true[l][..., :2] = (sigmoid(m_true[l][..., :2]) +
grid) / np.array(grid_shape[::-1])
m_true[l][..., 2:4] = np.exp(
m_true[l][..., 2:4]) * anchors_tensor / np.array(
input_shape[::-1])
m_true[l][..., 4] = sigmoid(m_true[l][..., 4])
m_true[l][..., 5:] = sigmoid(m_true[l][..., 5:])
#print("inside")
box = np.where(y_true[l][..., 4] > 0.5)
box = np.transpose(box)
for i in range(len(box)):
y_true[l][tuple(box[i])] = m_true[l][tuple(box[i])]
yield [image_data, *y_true], np.zeros(batch_size)
def data_generator(annotation_lines, batch_size, input_shape, anchors,
num_classes, model):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
while True:
image_data = []
box_data = []
for b in range(batch_size):
if i == 0:
np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i],
input_shape,
random=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 = preprocess_true_boxes(box_data, input_shape, anchors,
num_classes)
m_true = model.predict(image_data)
for l in range(len(m_true)):
#print(l)
m_true[l][..., :2] = sigmoid(m_true[l][..., :2])
m_true[l][..., 2:4] = np.exp(m_true[l][..., 2:4])
m_true[l][..., 4] = sigmoid(m_true[l][..., 4])
m_true[l][..., 5:] = sigmoid(m_true[l][..., 5:])
#print("inside")
#print(m_true[l].shape)
box = np.where(m_true[l][:, :, :, :, 4] > 0.3)
for i in range(len(box[0])):
s = np.array(box)
#print( "({},{},{},{})".format(s[0,i],s[1,i],s[2,i],s[3,i]) )
#print( m_true[l][s[0,i],s[1,i],s[2,i],s[3,i],0:5] )
#objprob = m_true[l][s[0,i],s[1,i],s[2,i],s[3,i],5:25]
#print( objprob )
#obnum = np.argmax( objprob )
y_true[l][s[0, i], s[1, i], s[2, i],
s[3, i]] = m_true[l][s[0, i], s[1, i], s[2, i], s[3,
i]]
#print("-------------------------------------------------------")
yield [image_data, *y_true] #, np.zeros(batch_size)
def data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
while True:
image_data = []
box_data = []
for b in range(batch_size):
if i==0:
np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i], input_shape, random=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 = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)
yield [image_data, *y_true], np.zeros(batch_size)
def data_generator(annotation_lines, batch_size, input_shape, anchors,
num_classes):
'''data generator for fit_generator'''
num_anchors = len(anchors)
image_input = Input(shape=(416, 416, 3))
model = yolo_body(image_input, num_anchors // 3, num_classes)
model.load_weights("model_data/trained_weights_final.h5")
n = len(annotation_lines)
i = 0
while True:
image_data = []
box_data = []
for b in range(batch_size):
if i == 0:
np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i],
input_shape,
random=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 = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)
# print(image_data.shape)
y_true = model.predict(image_data)
# print("d")
# print(y_true[0].shape)
# print(y_true[1].shape)
# print(y_true[2].shape)
# y_true[0] = y_true[0].reshape(y_true[0].shape[0], y_true[0].shape[1], y_true[0].shape[2], 3 , y_true[0].shape[3]//3 )
# y_true[1] = y_true[1].reshape(y_true[1].shape[0], y_true[1].shape[1], y_true[1].shape[2], 3 , y_true[1].shape[3]//3 )
# y_true[2] = y_true[2].reshape(y_true[2].shape[0], y_true[2].shape[1], y_true[2].shape[2], 3 , y_true[2].shape[3]//3 )
yield [image_data, *y_true] #, np.zeros(batch_size)
def data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes,teacher):
'''data generator for fit_generator'''
n = len(annotation_lines)
i = 0
while True:
image_data = []
box_data = []
for b in range(batch_size):
if i==0:
np.random.shuffle(annotation_lines)
image, box = get_random_data(annotation_lines[i], input_shape, random=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 = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)
m_true = teacher.predict(image_data)
h, w = input_shape
num_anchors = len(anchors)
l_true = [ np.zeros( shape=( batch_size ,416//{0:32, 1:16, 2:8}[l], 416//{0:32, 1:16, 2:8}[l], 9//3, 20+5) ) for l in range(3) ]
#print(len(y_true))
#print(len(m_true))
#print(len(l_true))
anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if len(m_true)==3 else [[3,4,5], [1,2,3]]
for l in range( len(m_true) ) :
'''
pred_output = tf.Variable(m_true[l])
anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if len(m_true)==3 else [[3,4,5], [1,2,3]]
pred_xy, pred_wh , pred_conf , pred_class = yolo_head( pred_output ,anchors[anchor_mask[l]], num_classes, input_shape, calc_loss=False)
pred_model = K.concatenate([pred_xy, pred_wh, pred_conf ,pred_class ])
with tf.Session() as sess:
init = tf.global_variables_initializer()
sess.run(init)
pred_model = pred_model.eval()
'''
anchors_tensor = np.reshape( anchors[anchor_mask[l]] , [1, 1, 1, len( anchors[anchor_mask[l]] ) , 2] )
grid_shape = m_true[l].shape[1:3] # height, width
grid_shape
grid_y = np.tile(np.reshape(np.arange(0, stop=grid_shape[0]), [-1, 1, 1, 1]),
[1, grid_shape[1], 1, 1])
grid_x = np.tile(np.reshape(np.arange(0, stop=grid_shape[1]), [1, -1, 1, 1]),
[grid_shape[0], 1, 1, 1])
grid = np.concatenate([grid_x, grid_y],axis=-1)
#print(l)
m_true[l][...,:2] = (sigmoid(m_true[l][...,:2]) + grid ) / np.array( grid_shape[::-1] )
m_true[l][..., 2:4] = np.exp(m_true[l][..., 2:4]) * anchors_tensor / np.array( input_shape[::-1] )
m_true[l][..., 4] = sigmoid(m_true[l][..., 4])
m_true[l][..., 5:] = sigmoid(m_true[l][..., 5:])
#print("inside")
box = np.where(y_true[l][...,4] > 0.5 )
box = np.transpose(box)
for i in range(len(box)):
l_true[l][tuple(box[i])] = m_true[l][tuple(box[i])] #pred_model[tuple(box[i])]
yield [image_data, *y_true , *l_true ], np.zeros(batch_size)
