soshink / convolutional-color-constancy

convolutional-color-constancy

Jonathan T. Barron; Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2015, pp. 379-387

convolutional-color-constancy's Issues

How to process the PNG16 files in Cube++ dataset?

Hi, thanks for sharing your implementation. I have a few of problems about the usage of the Cube++ dataset.

The PNG files provided in Cube++ are in 16-bits format. It seems that these images cannot be read correctly using cv::imread . I tried to use cv2.imread(img_name, cv2.IMREAD_UNCHANGED), but the images are still too dark.

Would you please share some details about the procesing the PNG16 files?

Thanks for your reading. I will be appreciated for your reply!

def __getitem__(self, idx):
        img_name = self.data_samples[idx]
        img = cv2.imread(str(img_name))
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB).astype(np.float64)

        if self.return_gt_rgb:
            if self.augmentations is None:
                return img, np.array(self.gt_csv[img_name.stem])
            else:
                return self.augmentations(img), np.array(self.gt_csv[img_name.stem])
        if self.augmentations is None:
            return img, rgb2uvy([self.gt_csv[img_name.stem]])[0, :2]
        else:
            return self.augmentations(img), rgb2uvy([self.gt_csv[img_name.stem]])[0, :2]