keras-image-super-resolution
keras-image-super-resolution copied to clipboard
pnsr - inf
Why do you use matlab func to generate dataset instead doing it directly in load functions of python?
I have generated dataset processing 800 to 100 (train / valid) cropped by 74x74 and with 6 different augmentation flipping for each tile. Collecting around 15000 image pairs for valid and 115000 image pairs for training. Correctly?
for each tile sized to 74 x 74 for method in range(6): if method == 0: image = np.rot90(image) elif method == 1: image = np.fliplr(image) elif method == 2: image = np.flipud(image) elif method == 3: image = np.rot90(np.rot90(image)) elif method == 4: image = np.rot90(np.fliplr(image)) elif method == 5: image = np.rot90(np.flipud(image))