metric_utils.py [167:214]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        self.size = size
        self.net = net
        self.results = []
        self.device = device if device is not None else torch.device(
            'cuda' if torch.cuda.is_available() else 'cpu')
        self.fn = lpips.LPIPS(net=net).eval().to(self.device)

    def measure(self):
        return np.mean(self.results)

    def report(self):
        return f'LPIPS ({self.net}) = {self.measure():.6f}'

    def read_img_list(self, img_list):
        size = self.size
        images = []
        white_background = np.ones((size, size, 3), dtype=np.uint8) * 255

        for img_path in img_list:
            img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)

            if img.shape[2] == 4:  # Handle BGRA images
                alpha = img[:, :, 3]  # Extract alpha channel
                img = cv2.cvtColor(img,
                                   cv2.COLOR_BGRA2BGR)  # Convert BGRA to BGR

                img = cv2.cvtColor(img,
                                   cv2.COLOR_BGR2RGB)  # Convert BGR to RGB
                img[np.where(alpha == 0)] = [
                    255, 255, 255
                ]  # Set transparent pixels to white
            else:  # Handle other image formats like JPG and PNG
                img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

            img = cv2.resize(img, (size, size), interpolation=cv2.INTER_AREA)
            images.append(img)

        images = np.stack(images, axis=0)
        # images[np.where(images == 0)] = 255  # Set black pixels to white
        # images = np.where(images == 0, white_background, images)  # Set transparent pixels to white
        images = images.astype(np.float32) / 255.0

        return images

    # * recommend to use this function for evaluation
    @torch.no_grad()
    def score_gt(self, ref_paths, novel_paths):
        self.results = []
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



metric_utils.py [249:296]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
        self.size = size
        self.net = net
        self.results = []
        self.device = device if device is not None else torch.device(
            'cuda' if torch.cuda.is_available() else 'cpu')
        self.fn = lpips.LPIPS(net=net).eval().to(self.device)

    def measure(self):
        return np.mean(self.results)

    def report(self):
        return f'LPIPS ({self.net}) = {self.measure():.6f}'

    def read_img_list(self, img_list):
        size = self.size
        images = []
        white_background = np.ones((size, size, 3), dtype=np.uint8) * 255

        for img_path in img_list:
            img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)

            if img.shape[2] == 4:  # Handle BGRA images
                alpha = img[:, :, 3]  # Extract alpha channel
                img = cv2.cvtColor(img,
                                   cv2.COLOR_BGRA2BGR)  # Convert BGRA to BGR

                img = cv2.cvtColor(img,
                                   cv2.COLOR_BGR2RGB)  # Convert BGR to RGB
                img[np.where(alpha == 0)] = [
                    255, 255, 255
                ]  # Set transparent pixels to white
            else:  # Handle other image formats like JPG and PNG
                img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

            img = cv2.resize(img, (size, size), interpolation=cv2.INTER_AREA)
            images.append(img)

        images = np.stack(images, axis=0)
        # images[np.where(images == 0)] = 255  # Set black pixels to white
        # images = np.where(images == 0, white_background, images)  # Set transparent pixels to white
        images = images.astype(np.float32) / 255.0

        return images

    # * recommend to use this function for evaluation
    @torch.no_grad()
    def score_gt(self, ref_paths, novel_paths):
        self.results = []
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -