def __init__(self, input_data, trainable):
self.trainable = trainable
self.classes = utils.read_class_names(cfg.YOLO.CLASSES)
self.num_class = len(self.classes)
self.strides = np.array(cfg.YOLO.STRIDES)
self.anchors = utils.get_anchors(cfg.YOLO.ANCHORS)
self.anchor_per_scale = cfg.YOLO.ANCHOR_PER_SCALE
self.iou_loss_thresh = cfg.YOLO.IOU_LOSS_THRESH
self.upsample_method = cfg.YOLO.UPSAMPLE_METHOD
self.layer_nums = cfg.YOLO.LAYER_NUMS
try:
self.conv_lbbox, self.conv_mbbox, self.out = self.__build_nework(
input_data)
except:
raise NotImplementedError("Can not build up yolov3 network!")
print(self.out)
self.predict_op = tf.argmax(input=self.out,
axis=1,
name='layer_classes')
print("layer_nums::", self.predict_op)
with tf.variable_scope('pred_sbbox'):
self.pred_lbbox = self.decode(self.conv_lbbox, self.anchors[0],
self.strides[0])
with tf.variable_scope('pred_mbbox'):
self.pred_mbbox = self.decode(self.conv_mbbox, self.anchors[1],
self.strides[1])
def __init__(self):
self.input_size = cfg.TEST.INPUT_SIZE
self.anchor_per_scale = cfg.YOLO.ANCHOR_PER_SCALE
self.classes = utils.read_class_names(cfg.YOLO.CLASSES)
self.num_classes = len(self.classes)
self.anchors = np.array(utils.get_anchors(cfg.YOLO.ANCHORS))
self.score_threshold = cfg.TEST.SCORE_THRESHOLD
self.iou_threshold = cfg.TEST.IOU_THRESHOLD
self.moving_ave_decay = cfg.YOLO.MOVING_AVE_DECAY
self.annotation_path = cfg.TEST.ANNOT_PATH
self.weight_file = cfg.TEST.WEIGHT_FILE
self.write_image = cfg.TEST.WRITE_IMAGE
self.write_image_path = cfg.TEST.WRITE_IMAGE_PATH
self.show_label = cfg.TEST.SHOW_LABEL
with tf.name_scope('input'):
self.input_data = tf.placeholder(dtype=tf.float32,
name='input_data')
self.trainable = tf.placeholder(dtype=tf.bool, name='trainable')
model = YOLOV3(self.input_data, self.trainable)
self.pred_sbbox, self.pred_mbbox, self.pred_lbbox = model.pred_sbbox, model.pred_mbbox, model.pred_lbbox
#
# with tf.name_scope('ema'):
# ema_obj = tf.train.ExponentialMovingAverage(self.moving_ave_decay)
self.sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True))
# self.saver = tf.train.Saver(ema_obj.variables_to_restore())
self.saver = tf.train.Saver()
self.saver.restore(self.sess, self.weight_file)
def __init__(self, dataset_type):
self.annot_path = cfg.TRAIN.ANNOT_PATH if dataset_type == 'train' else cfg.TEST.ANNOT_PATH
self.input_sizes = cfg.TRAIN.INPUT_SIZE if dataset_type == 'train' else cfg.TEST.INPUT_SIZE
self.batch_size = cfg.TRAIN.BATCH_SIZE if dataset_type == 'train' else cfg.TEST.BATCH_SIZE
self.batch_size = self.batch_size
self.data_aug = cfg.TRAIN.DATA_AUG if dataset_type == 'train' else cfg.TEST.DATA_AUG
self.train_input_size = cfg.TRAIN.INPUT_SIZE
self.strides = np.array(cfg.YOLO.STRIDES)
self.classes = utils.read_class_names(cfg.YOLO.CLASSES)
self.num_classes = len(self.classes)
self.anchors = np.array(utils.get_anchors(cfg.YOLO.ANCHORS))
self.anchor_per_scale = cfg.YOLO.ANCHOR_PER_SCALE
self.max_bbox_per_scale = 150
self.annotations = self.load_annotations(dataset_type)
self.num_samples = len(self.annotations)
self.num_batchs = int(np.ceil(self.num_samples / self.batch_size))
self.batch_count = 0
def __init__(self):
self.input_size = cfg.TEST.INPUT_SIZE
self.anchor_per_scale = cfg.YOLO.ANCHOR_PER_SCALE
self.classes = utils.read_class_names(cfg.YOLO.CLASSES)
self.num_classes = len(self.classes)
self.anchors = np.array(utils.get_anchors(cfg.YOLO.ANCHORS))
self.score_threshold = cfg.TEST.SCORE_THRESHOLD
self.iou_threshold = cfg.TEST.IOU_THRESHOLD
self.moving_ave_decay = cfg.YOLO.MOVING_AVE_DECAY
self.annotation_path = cfg.TEST.ANNOT_PATH
self.weight_file = cfg.TEST.WEIGHT_FILE
self.write_image = cfg.TEST.WRITE_IMAGE
self.write_image_path = cfg.TEST.WRITE_IMAGE_PATH
self.show_label = cfg.TEST.SHOW_LABEL
self.pb_file = "E:/ckpt_dirs/Food_detection/multi_food/20190910/yolo_model.pb"
graph = tf.Graph()
with graph.as_default():
output_graph_def = tf.GraphDef()
with open(self.pb_file, "rb") as f:
output_graph_def.ParseFromString(f.read())
_ = tf.import_graph_def(output_graph_def, name="")
self.sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True))
self.input = self.sess.graph.get_tensor_by_name(
"define_input/input_data:0")
self.trainable = self.sess.graph.get_tensor_by_name(
"define_input/training:0")
self.pred_sbbox = self.sess.graph.get_tensor_by_name(
"define_loss/pred_sbbox/concat_2:0")
self.pred_mbbox = self.sess.graph.get_tensor_by_name(
"define_loss/pred_mbbox/concat_2:0")
self.pred_lbbox = self.sess.graph.get_tensor_by_name(
"define_loss/pred_lbbox/concat_2:0")
self.layer_num = graph.get_tensor_by_name(
"define_loss/layer_classes:0")
def __init__(self):
self.input_size = cfg.TEST.INPUT_SIZE
self.anchor_per_scale = cfg.YOLO.ANCHOR_PER_SCALE
self.classes = utils.read_class_names(cfg.YOLO.CLASSES)
self.num_classes = len(self.classes)
self.anchors = np.array(utils.get_anchors(cfg.YOLO.ANCHORS))
self.score_threshold = cfg.TEST.SCORE_THRESHOLD
self.iou_threshold = cfg.TEST.IOU_THRESHOLD
self.moving_ave_decay = cfg.YOLO.MOVING_AVE_DECAY
self.annotation_path = cfg.TEST.ANNOT_PATH
self.weight_file = "E:/ckpt_dirs/Food_detection/multi_food3/checkpoint/yolov3_train_loss=11.4018.ckpt-98"
self.write_image = cfg.TEST.WRITE_IMAGE
self.write_image_path = cfg.TEST.WRITE_IMAGE_PATH
self.show_label = cfg.TEST.SHOW_LABEL
graph = tf.Graph()
with graph.as_default():
self.saver = tf.train.import_meta_graph("{}.meta".format(
self.weight_file))
self.sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True))
self.saver.restore(self.sess, self.weight_file)
self.input = graph.get_tensor_by_name("define_input/input_data:0")
self.trainable = graph.get_tensor_by_name(
"define_input/training:0")
self.pred_sbbox = graph.get_tensor_by_name(
"define_loss/pred_sbbox/concat_2:0")
self.pred_mbbox = graph.get_tensor_by_name(
"define_loss/pred_mbbox/concat_2:0")
self.pred_lbbox = graph.get_tensor_by_name(
"define_loss/pred_lbbox/concat_2:0")
self.layer_num = graph.get_tensor_by_name(
"define_loss/layer_classes:0")
def __init__(self):
self.anchor_per_scale = cfg.YOLO.ANCHOR_PER_SCALE
self.classes = utils.read_class_names(cfg.YOLO.CLASSES)
self.num_classes = len(self.classes)
self.learn_rate_init = cfg.TRAIN.LEARN_RATE_INIT
self.learn_rate_end = cfg.TRAIN.LEARN_RATE_END
self.first_stage_epochs = cfg.TRAIN.FISRT_STAGE_EPOCHS
self.second_stage_epochs = cfg.TRAIN.SECOND_STAGE_EPOCHS
self.warmup_periods = cfg.TRAIN.WARMUP_EPOCHS
self.initial_weight = cfg.TRAIN.INITIAL_WEIGHT
self.time = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))
self.moving_ave_decay = cfg.YOLO.MOVING_AVE_DECAY
self.train_input_size = cfg.TRAIN.INPUT_SIZE
self.max_bbox_per_scale = 150
self.train_logdir = "./data/log/train"
self.trainset = Dataset('train')
self.testset = Dataset('test')
self.steps_per_period = len(self.trainset)
self.sess = tf.Session(config=config)
with tf.name_scope('define_input'):
self.input_data = tf.placeholder(dtype=tf.float32,
shape=(None, self.train_input_size, self.train_input_size, 3),
name='input_data')
self.layer_label = tf.placeholder(dtype=tf.float32, name='layer_label')
self.label_sbbox = tf.placeholder(dtype=tf.float32, name='label_sbbox')
self.label_mbbox = tf.placeholder(dtype=tf.float32, name='label_mbbox')
self.label_lbbox = tf.placeholder(dtype=tf.float32, name='label_lbbox')
self.true_sbboxes = tf.placeholder(dtype=tf.float32, name='sbboxes')
self.true_mbboxes = tf.placeholder(dtype=tf.float32, name='mbboxes')
self.true_lbboxes = tf.placeholder(dtype=tf.float32, name='lbboxes')
self.trainable = tf.placeholder(dtype=tf.bool, shape=[], name='training')
with tf.name_scope("define_loss"):
self.model = YOLOV3(self.input_data, self.trainable)
self.net_var = tf.global_variables()
self.giou_loss, self.conf_loss, self.prob_loss= self.model.compute_loss(
self.label_sbbox, self.label_mbbox, self.label_lbbox,
self.true_sbboxes, self.true_mbboxes, self.true_lbboxes)
self.layer_loss = self.model.layer_loss(self.layer_label)
print("layer_loss::::")
print(self.layer_loss)
self.l2_loss = tf.add_n([tf.nn.l2_loss(var) for var in tf.trainable_variables()])
self.layer_loss = tf.cond(self.layer_loss > 0.01, lambda: self.layer_loss, lambda: 0.0)
self.loss = self.giou_loss + self.conf_loss + 2 * self.prob_loss + 10 * self.layer_loss + 1e-5 * self.l2_loss
self.layer_out = self.model.out
with tf.name_scope('learn_rate'):
self.global_step = tf.Variable(1.0, dtype=tf.float64, trainable=False, name='global_step')
# self.learn_rate = tf.train.exponential_decay(self.learn_rate_init, global_step=self.global_step,
# decay_steps=1000, decay_rate=0.9)
# self.learn_rate = tf.maximum(self.learn_rate, self.learn_rate_end)
warmup_steps = tf.constant(self.warmup_periods * self.steps_per_period,
dtype=tf.float64, name='warmup_steps')
train_steps = tf.constant((self.first_stage_epochs + self.second_stage_epochs) * self.steps_per_period,
dtype=tf.float64, name='train_steps')
self.learn_rate = tf.cond(
pred=self.global_step < warmup_steps,
true_fn=lambda: self.global_step / warmup_steps * self.learn_rate_init,
false_fn=lambda: self.learn_rate_end + 0.5 * (self.learn_rate_init - self.learn_rate_end) *
(1 + tf.cos(
(self.global_step - warmup_steps) / (train_steps - warmup_steps) * np.pi))
)
global_step_update = tf.assign_add(self.global_step, 1.0)
with tf.name_scope("define_weight_decay"):
moving_ave = tf.train.ExponentialMovingAverage(self.moving_ave_decay).apply(tf.trainable_variables())
with tf.name_scope("define_first_stage_train"):
self.first_stage_trainable_var_list = []
for var in tf.trainable_variables():
var_name = var.op.name
var_name_mess = str(var_name).split('/')
if var_name_mess[0] in ['conv_sbbox', 'conv_mbbox', 'conv_lbbox']:
self.first_stage_trainable_var_list.append(var)
first_stage_optimizer = tf.train.AdamOptimizer(self.learn_rate).minimize(self.loss,
var_list=self.first_stage_trainable_var_list)
# first_stage_optimizer = tf.train.AdamOptimizer(self.learn_rate).minimize(self.loss)
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
with tf.control_dependencies([first_stage_optimizer, global_step_update]):
with tf.control_dependencies([moving_ave]):
self.train_op_with_frozen_variables = tf.no_op()
with tf.name_scope("define_second_stage_train"):
second_stage_trainable_var_list = tf.trainable_variables()
second_stage_optimizer = tf.train.AdamOptimizer(self.learn_rate).minimize(self.loss,
var_list=second_stage_trainable_var_list)
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
with tf.control_dependencies([second_stage_optimizer, global_step_update]):
with tf.control_dependencies([moving_ave]):
self.train_op_with_all_variables = tf.no_op()
with tf.name_scope('loader_and_saver'):
self.loader = tf.train.Saver(self.net_var)
self.saver = tf.train.Saver(tf.global_variables(), max_to_keep=10)
with tf.name_scope('summary'):
tf.summary.scalar("learn_rate", self.learn_rate)
tf.summary.scalar("giou_loss", self.giou_loss)
tf.summary.scalar("conf_loss", self.conf_loss)
tf.summary.scalar("prob_loss", self.prob_loss)
tf.summary.scalar("layer_loss", self.layer_loss)
tf.summary.scalar("total_loss", self.loss)
test_logdir = "./data/log/test"
if os.path.exists(test_logdir): shutil.rmtree(test_logdir)
os.mkdir(test_logdir)
if os.path.exists(self.train_logdir): shutil.rmtree(self.train_logdir)
os.mkdir(self.train_logdir)
self.write_op = tf.summary.merge_all()
self.train_summary_writer = tf.summary.FileWriter(self.train_logdir, graph=self.sess.graph)
self.test_summary_writer = tf.summary.FileWriter(test_logdir, graph=self.sess.graph)
best_bboxes = self.get_top_cls(pred_bbox, org_h, org_w, self.top_n) # 获取top_n类别和置信度
bboxes = utils.postprocess_boxes(pred_bbox, (org_h, org_w), self.input_size, self.score_threshold)
bboxes = utils.nms(bboxes, self.iou_threshold)
layer_n = layer_[0] # 烤层结果
return bboxes, layer_n, best_bboxes
if __name__ == '__main__':
img_root = "F:/serve_data/202101-04/covert_jpg" # 图片地址
img_save = "F:/serve_data/202101-04/classes" # 图片地址
import time
classes = utils.read_class_names(cfg.YOLO.CLASSES)
classes[40] = "potatom"
classes[41] = "sweetpotatom"
classes[101] = "chiffon_size4"
start_time = time.time()
Y = YoloPredic()
end_time0 = time.time()
print("加载时间：", end_time0 - start_time)
for img in tqdm(os.listdir(img_root)):
img_path = img_root + "/" + img
image = cv2.imread(img_path) # 图片读取
bboxes, layer_n, best_bboxes = Y.predict(image)
bboxes, layer_n, best_bboxes = correct_bboxes(bboxes, layer_n, best_bboxes) # 矫正输出结果
if len(bboxes) > 0:
def __init__(self):
self.anchor_per_scale = cfg.YOLO.ANCHOR_PER_SCALE
self.classes = utils.read_class_names(cfg.YOLO.CLASSES)
self.num_classes = len(self.classes)
self.learn_rate_init = cfg.TRAIN.LEARN_RATE_INIT
self.learn_rate_end = cfg.TRAIN.LEARN_RATE_END
self.first_stage_epochs = cfg.TRAIN.FISRT_STAGE_EPOCHS
self.second_stage_epochs = cfg.TRAIN.SECOND_STAGE_EPOCHS
self.warmup_periods = cfg.TRAIN.WARMUP_EPOCHS
self.initial_weight = cfg.TRAIN.INITIAL_WEIGHT
self.time = time.strftime('%Y-%m-%d-%H-%M-%S',
time.localtime(time.time()))
self.moving_ave_decay = cfg.YOLO.MOVING_AVE_DECAY
self.train_input_size = cfg.TRAIN.INPUT_SIZE
self.max_bbox_per_scale = 150
self.train_logdir = "./data/log/train"
self.trainset = Dataset('train')
self.testset = Dataset('test')
self.steps_per_period = len(self.trainset)
self.sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True))
self.tower_grads = []
with tf.name_scope('define_input'):
self.input_data = tf.placeholder(dtype=tf.float32,
shape=(None,
self.train_input_size,
self.train_input_size, 3),
name='input_data')
self.layer_label = tf.placeholder(dtype=tf.float32,
name='layer_label')
self.label_sbbox = tf.placeholder(dtype=tf.float32,
name='label_sbbox')
self.label_mbbox = tf.placeholder(dtype=tf.float32,
name='label_mbbox')
self.label_lbbox = tf.placeholder(dtype=tf.float32,
name='label_lbbox')
self.true_sbboxes = tf.placeholder(dtype=tf.float32,
name='sbboxes')
self.true_mbboxes = tf.placeholder(dtype=tf.float32,
name='mbboxes')
self.true_lbboxes = tf.placeholder(dtype=tf.float32,
name='lbboxes')
self.trainable = tf.placeholder(dtype=tf.bool, name='training')
with tf.name_scope('learn_rate'):
self.global_step = tf.Variable(1.0,
dtype=tf.float64,
trainable=False,
name='global_step')
self.learn_rate = tf.train.exponential_decay(
self.learn_rate_init,
global_step=self.global_step,
decay_steps=1000,
decay_rate=0.9)
self.learn_rate = tf.maximum(self.learn_rate, self.learn_rate_end)
global_step_update = tf.assign_add(self.global_step, 1.0)
with tf.variable_scope(tf.get_variable_scope(), reuse=tf.AUTO_REUSE):
self.opt = tf.train.GradientDescentOptimizer(self.learn_rate)
with tf.name_scope(
"define_loss",
tf.variable_scope("define_loss", reuse=tf.AUTO_REUSE)):
for i in range(4):
with tf.device('/gpu:{}'.format(i + 4)):
self.input_data_i = self.input_data[i *
cfg.TRAIN.BATCH_SIZE:
(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.layer_label_i = self.layer_label[i *
cfg.TRAIN.BATCH_SIZE:
(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.label_sbbox_i = self.label_sbbox[i *
cfg.TRAIN.BATCH_SIZE:
(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.label_mbbox_i = self.label_mbbox[i *
cfg.TRAIN.BATCH_SIZE:
(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.label_lbbox_i = self.label_lbbox[i *
cfg.TRAIN.BATCH_SIZE:
(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.true_sbboxes_i = self.true_sbboxes[
i * cfg.TRAIN.BATCH_SIZE:(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.true_mbboxes_i = self.true_mbboxes[
i * cfg.TRAIN.BATCH_SIZE:(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.true_lbboxes_i = self.true_lbboxes[
i * cfg.TRAIN.BATCH_SIZE:(i + 1) *
cfg.TRAIN.BATCH_SIZE]
self.model = YOLOV3(self.input_data_i, self.trainable)
self.layer_out = self.model.out
self.net_var = tf.global_variables()
self.giou_loss, self.conf_loss, self.prob_loss, self.giou, self.bbox_loss_scale = self.model.compute_loss(
self.label_sbbox_i, self.label_mbbox_i,
self.label_lbbox_i, self.true_sbboxes_i,
self.true_mbboxes_i, self.true_lbboxes_i)
# self.layer_loss = self.model.layer_loss(self.layer_label_i)
# self.layer_loss = tf.cond(self.layer_loss > 0.01, lambda: self.layer_loss, lambda: 0.0)
self.layer_loss = 0
self.loss = self.giou_loss + self.conf_loss + 2 * self.prob_loss + 10 * self.layer_loss
tf.get_variable_scope().reuse_variables()
grads = self.opt.compute_gradients(self.loss)
self.tower_grads.append(grads)
grads = average_gradients(self.tower_grads)
self.train_op = self.opt.apply_gradients(grads)
with tf.name_scope('loader_and_saver'):
self.loader = tf.train.Saver(self.net_var)
self.saver = tf.train.Saver(tf.global_variables(), max_to_keep=10)
with tf.name_scope('summary'):
tf.summary.scalar("learn_rate", self.learn_rate)
tf.summary.scalar("giou_loss", self.giou_loss)
tf.summary.scalar("conf_loss", self.conf_loss)
tf.summary.scalar("prob_loss", self.prob_loss)
tf.summary.scalar("layer_loss", self.layer_loss)
tf.summary.scalar("total_loss", self.loss)
test_logdir = "./data/log/test"
if os.path.exists(test_logdir): shutil.rmtree(test_logdir)
os.mkdir(test_logdir)
if os.path.exists(self.train_logdir):
shutil.rmtree(self.train_logdir)
os.mkdir(self.train_logdir)
self.write_op = tf.summary.merge_all()
self.train_summary_writer = tf.summary.FileWriter(
self.train_logdir, graph=self.sess.graph)
self.test_summary_writer = tf.summary.FileWriter(
test_logdir, graph=self.sess.graph)
# 参数量计算
flops = tf.profiler.profile(
self.sess.graph,
options=tf.profiler.ProfileOptionBuilder.float_operation())
params = tf.profiler.profile(self.sess.graph,
options=tf.profiler.ProfileOptionBuilder.
trainable_variables_parameter())
print('FLOPs: {}; Trainable params: {}'.format(
flops.total_float_ops, params.total_parameters))