def __init__(self):
super().__init__()
self.image = None
print_log("Start", "debug")
self.initUI()
print_log("End", "debug")
def batch(self, image):
print_log("Start")
batch = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])(image).unsqueeze(0)
self._batch = batch
print_log("End")
def get_classes(self):
sum = self.predictions.shape[0] * self.predictions.shape[1]
classes, counts = np.unique(self.predictions, return_counts=True)
classes = [Label(_class).name for _class in classes]
percentages = [count / sum * 100 for count in counts]
print_log("There are {} classes.".format(len(classes)))
for _class, percentage in zip(classes, percentages):
print_log("'{}' ({:.2f}%)".format(_class, percentage))
return classes, percentages
def to_cuda(self):
if self.cuda:
print_log("using cuda")
self.batch.to('cuda')
self.model.to('cuda')
else:
print_log("using cpu")
self.batch.to('cpu')
self.model.to('cpu')
return self.cuda
def show_image(self, img):
print_log("Start")
qimage = None
if isinstance(img, str):
qimage = QImage(img)
elif isinstance(img, np.ndarray):
qimage = q2n.array2qimage(img, normalize=False)
pixmap = QPixmap.fromImage(qimage)
im_w, im_h = pixmap.size().width(), pixmap.size().height()
if im_w > im_h:
pixmap = pixmap.scaledToWidth(self.label_min)
else:
pixmap = pixmap.scaledToHeight(self.label_min)
self.imageLabel.setPixmap(pixmap)
self.imageLabel.setAlignment(Qt.AlignCenter)
self.image = cv2.imread(img)
cv2.cvtColor(self.image, cv2.COLOR_BGR2RGB)
print_log("End")
def detect_image(self):
print_log("Start")
self.set_status("Detecting...")
if self.image is None:
print_log("Image not loaded!", "warn")
self.set_status("You must load an image.")
self.set_status("Detecting finished.")
print_log("End")
def model(self, model_name="deeplabv3_resnet101"):
print_log("Start")
print_log("model: {}".format(model_name))
_model = torch.hub.load('pytorch/vision:v0.6.0',
model_name,
pretrained=True)
_model.eval()
print_log("End")
self._model = _model
def inference(self):
print_log("Start")
start_t = time.time()
with torch.no_grad():
self._output = self.model(self.batch)['out'][0]
end_t = time.time()
t = end_t - start_t
print_log("Inference time: {}".format(
time.strftime("%Mm %S.%fs", time.gmtime(t))[:-3]))
self._predictions = self.output.argmax(0).byte().cpu().numpy()
print_log("End")
def image(self, image):
print_log("Start")
print_log("Type: {}".format(type(image)))
if isinstance(image, str):
image = cv2.imread(image)
self._image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
elif isinstance(image, np.ndarray):
self._image = image
elif isinstance(image, Image.Image):
self._image = np.asarray(image)
print_log("Shape: {}".format(self._image.shape))
def get_prediction(self, label: str):
print_log("class: {}".format(label))
prediction = np.where(self.predictions == Label[label.upper()].value,
1, 0)
return prediction
def initUI(self):
print_log("Start", "debug")
self.setFixedSize(WIDTH, HEIGHT)
self.setWindowTitle('Scroll Area Demonstration')
# Widgets
self.imageLabel = QLabel("ImageLabel")
self.imageLabel.setBackgroundRole(QPalette.Dark)
self.imageLabel = QLabel("ImageLabel")
label_w, label_h = self.imageLabel.size().width(
), self.imageLabel.size().height()
self.label_min = min(label_w, label_h)
self.imageLabel.resize(self.label_min, self.label_min)
print_log("Resize imageLabel")
self.show_image(os.path.join(IMG_DIR, "sample2.jpg"))
self.set_status("HI")
self.openBtn = QPushButton("Open")
self.openBtn.clicked.connect(self.open_image)
self.detectBtn = QPushButton("Detect")
self.detectBtn.clicked.connect(self.detect_image)
self.compareBtn = QPushButton("Compare")
self.viewWideBtn = QPushButton("View Wide")
self.saveBtn = QPushButton("Save")
self.scrollArea = QScrollArea()
# Layouts
self.btnGroup = QGridLayout()
self.btnGroup.addWidget(self.openBtn, 0, 0)
self.btnGroup.addWidget(self.detectBtn, 0, 1)
self.btnGroup.addWidget(self.compareBtn, 0, 2)
self.btnGroup.addWidget(self.viewWideBtn, 0, 3)
self.btnGroup.addWidget(self.saveBtn, 0, 4)
self.viewLayout = QVBoxLayout()
self.layout = QVBoxLayout()
self.layout.addWidget(self.imageLabel)
self.layout.addLayout(self.btnGroup)
self.layout.addWidget(self.scrollArea)
# Shortcuts
self.openFile = QShortcut(QKeySequence("Ctrl+O"), self)
self.openFile.activated.connect(self.open_image)
centralWidget = QWidget()
centralWidget.setLayout(self.layout)
self.setCentralWidget(centralWidget)
print_log("End", "debug")
def open_image(self):
print_log("Start")
self.set_status("Select image...")
fname = QFileDialog.getOpenFileName(
self, 'Open image', str(Path.home()),
"Image files (*.jpg *.jpeg *.gif *.png)")
if fname[0]:
print_log("File path is '{}'".format(fname[0]))
try:
kind = filetype.guess(fname[0])
print_log("Mime type: '{}'".format(kind.mime))
if kind.mime.split("/")[0] == "image":
self.show_image(fname[0])
self.set_status("Opened.")
else:
print_log("Not Supported File")
self.set_status("Not Supported file..")
except FileNotFoundError as e:
print_log("File Not Found!")
self.set_status("File not found!")
else:
print_log("File path is not defined.")
self.set_status("File not selected.")
print_log("Finish")