def process_img(image: numpy.ndarray):
# Подготавливается конфигурация
config = PredictionConfig()
config.display()
# Производится определение модели
model = MaskRCNN(mode='inference', model_dir='./', config=config)
# Производится загрузка весов
model.load_weights(
WEIGHT_FILE,
by_name=True
)
# Производим сопоставление имен идентификаторов(Поскольку обучаем на части объектов)
objs_ids = dict()
objs_names = dict()
for i, class_id in enumerate(CLASS_IDS, start=1):
# print(i)
objs_ids[i] = int(OBJECTS[class_id].split('_')[0])
objs_names[i] = OBJECTS[class_id]
print(objs_ids)
print(objs_names)
identified_objects = list()
r = model.detect([image], verbose=0)[0]
# Если на кадре найдены объекты, то проводим их обработку
if len(r['class_ids']):
for num, object_id in enumerate(r['class_ids']):
box = r['rois'][num]
identified_object = IdentifiedObject(
class_id=objs_ids[object_id],
name=objs_names[object_id],
mask=r['masks'][:, :, num],
box=Box(box[1], box[0], box[3], box[2]),
score=r['scores'][num]
)
print(identified_object.__dict__)
identified_objects.append(identified_object)
return identified_objects
def train():
config = DeepFashionTrainConfig()
config.display()
model = MaskRCNN(mode="training", config=config, model_dir="checkpoints")
# Load coco pretrained weights
coco_weights = "models/mask_rcnn_coco.h5"
# As we don't want to detect default coco classes, we want to exclude the last layers.
# If not model requires matching number of coco classes
model.load_weights(coco_weights,
by_name=True,
exclude=[
"mrcnn_class_logits", "mrcnn_bbox_fc", "mrcnn_bbox",
"mrcnn_mask"
])
dataset_train = DeepFashionDataset()
dataset_train.load_coco(config.TRAIN_IMG_DIR,
config.TRAIN_ANNOTATIONS_PATH)
dataset_train.prepare()
dataset_valid = DeepFashionDataset()
dataset_valid.load_coco(config.VALID_IMG_DIR,
config.VALID_ANNOTATIONS_PATH)
dataset_valid.prepare()
model.train(dataset_train,
dataset_valid,
learning_rate=config.LEARNING_RATE,
epochs=30,
layers='heads')
print("Finish")
def instance_segment(image_path, model_path):
# define the model
rcnn = MaskRCNN(mode='inference',
model_dir='mask_rcnn',
config=TestConfig())
# load coco model weights
# Epochs 30
# Size (3000 train and 30000 validation)\
# Steps Per Epoch 1000
rcnn.load_weights(model_path, by_name=True)
# load photograph
img = load_img(image_path)
img = img_to_array(img)
# make prediction
results = rcnn.detect([img], verbose=0)
# get dictionary for first prediction
r = results[0]
# show photo with bounding boxes, masks, class labels and scores
display_instances(img, r['rois'], r['masks'], r['class_ids'], class_names,
r['scores'])
return r
app.config['JSON_SORT_KEYS'] = False
app.config["IMAGE_UPLOADS"] = "images"
# define the test configuration
class TestConfig(Config):
NAME = "test"
GPU_COUNT = 1
IMAGES_PER_GPU = 1
NUM_CLASSES = 1 + 13
model_path = '../models/mask_rcnn_deep_fashion_0050.h5'
global rcnn
rcnn = MaskRCNN(mode='inference',
model_dir='../mask_rcnn',
config=TestConfig())
rcnn.load_weights(model_path, by_name=True)
global graph
graph = tf.get_default_graph()
@app.route("/detect", methods=["POST"])
# @cross_origin()
def upload_image():
if request.files:
image = request.files["image"]
# load photograph
img = cv2.imdecode(np.frombuffer(image.read(), dtype=np.uint8), -1)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
result = extract_attributes(img)
config_proto = ConfigProto()
# config_proto.gpu_options.per_process_gpu_memory_fraction = 0.8
config_proto.gpu_options.allow_growth = True
session = InteractiveSession(config=config_proto)
# Подготавливаем тренировочный и тестовый набор данных
# train_set = DataSetFactory.new_instance('data/img')
# test_set = DataSetFactory.new_instance('data/val')
train_set = CocoSetFactory.new_instance(COCO_PATH, 'train', 2017, CLASS_IDS)
test_set = CocoSetFactory.new_instance(COCO_PATH, 'val', 2017, CLASS_IDS)
# Определяется конфигурация
config = DetectConfig()
config.display()
# Определяется модель
model = MaskRCNN(mode='training', model_dir='./', config=config)
# Произвести загрузку стартовой модели
model.load_weights(START_MODEL_PATH,
by_name=True,
exclude=[
"mrcnn_class_logits", "mrcnn_bbox_fc", "mrcnn_bbox",
"mrcnn_mask"
])
# augmentation = imgaug.augmenters.Fliplr(0.5)
# Запуск тренировки тестовой модели
model.train(train_set,
test_set,
learning_rate=config.LEARNING_RATE,
epochs=EPOCHS,
rect = Rectangle((x1, y1), width, height, fill=False, color='red')
# draw the box
ax.add_patch(rect)
# show the figure
pyplot.show()
# train_set = CocoSetFactory.new_instance(COCO_PATH, 'train', 2017, CLASS_IDS)
# test_set = CocoSetFactory.new_instance(COCO_PATH, 'val', 2017, CLASS_IDS)
train_set = VGGCocoSetFactory.new_instance(COCO_PATH, 'train', 1004, CLASS_IDS)
test_set = VGGCocoSetFactory.new_instance(COCO_PATH, 'val', 1004, CLASS_IDS)
# create config
cfg = DetectConfig()
# define the model
model = MaskRCNN(mode='inference', model_dir='./', config=cfg)
# load model weights
model.load_weights(weight_path, by_name=True)
# evaluate model on training dataset
print(
f"Start evaluate model train :{datetime.now()} weight_file:{weight_path}")
train_mAP = evaluate_model(train_set, model, cfg)
print("Train mAP: %.3f" % train_mAP)
# evaluate model on test dataset
print(f"Start evaluate model test:{datetime.now()}")
test_mAP = evaluate_model(test_set, model, cfg)
print("Test mAP: %.3f" % test_mAP)
print(f"End:{datetime.now()}")
# # plot predictions for train dataset
# WEIGHT_FILE = 'detect_cfg20200513T1306/mask_rcnn_detect_cfg_0080.h5'
# WEIGHT_FILE = 'mask_rcnn_detect_cfg_0082.h5'
# WEIGHT_FILE = 'detect_cfg20200618T1641/mask_rcnn_detect_cfg_0052.h5'
WEIGHT_FILE = 'detect_cfg20200618T1641/mask_rcnn_detect_cfg_0099.h5'
# VIDEO_PATH = 'flights1.mov'
# VIDEO_PATH = 'test/test.mp4'
# VIDEO_PATH = 'video/профессия нефтяник.mp4'
# VIDEO_PATH = 'video/test.mp4'
# VIDEO_PATH = 'main_test_video.mp4'
VIDEO_PATH = 'test_video/6.mp4'
OUTPUT_FILE = "result/result_{:%Y%m%dT%H%M%S}.avi".format(
datetime.datetime.now())
# Производится определение параметров использования видеокарты
define_video_config()
# Подготавливается конфигурация
config = PredictionConfig()
config.display()
# Производится определение модели
model = MaskRCNN(mode='inference', model_dir='./', config=config)
# Производится загрузка весов
model.load_weights(WEIGHT_FILE, by_name=True)
# Определяется помошник для работы с видео
video_helper = VideoHelper(model, VIDEO_PATH, OUTPUT_FILE)
# Производится запуск
video_helper.run()