def load_transform_image( item, root: Path, image_transform: Callable, debug: bool = False): image = load_image(item, root) image = image_transform(image) if debug: image.save('_debug.png') return tensor_transform(image)
def load_transform_image( image_path: Path, image_transform: Callable, debug: bool = False): image = cv2.imread(str(image_path.absolute())) image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) image = image_transform(image=image)["image"] # if debug: # image.save('_debug.jpg') tensor = tensor_transform(image=image)["image"] return tensor
def __getitem__(self, idx): item = self.df.iloc[idx % len(self.df)] image = load_image(item, self.root) height, width = image.shape[:2] if self.image_transform is not None: data = self.image_transform(image=image) image = data['image'] image = cv2.resize(image, (SIZE, SIZE)) image = tensor_transform(image) return image, item.id
def __getitem__(self, idx: int): item = self.df.iloc[idx] image = load_image(item, self.root) height, width = image.shape[:2] if self.image_transform is not None: data = self.image_transform(image=image) image = data['image'] # resize and transform to tensor image = cv2.resize(image, (SIZE, SIZE)) image = tensor_transform(image) # labels encoding target = torch.zeros(N_CLASSES) for cls in item.attribute_ids.split(): target[int(cls)] = int(1) return image, target
def load_transform_image( item, root: Path, image_transform: Callable, debug: bool = False,albumentation=True): image = load_image(item, root) if albumentation: image = np.array(image) data = {"image":image} image = image_transform(**data) image = image["image"] # albumentationした後もdebug=Trueの際に、png画像を保存できるようにする else: # pytorch 標準の場合 image = image_transform(image) if debug: image.save('_debug.png') # Pdb().set_trace() return tensor_transform(image)