How can i load image correctly using Data Generator with preprocessing

im trying to apply CLAHE to my dataset, but Im not sure if it is applied correctly. Im trying to visualize the result after applying CLAHE, but I can only visualize if I use plt.imshow(img[0].astype('uint8')). So I am not sure if it is being applied correctly. Ill be very grateful if someone can help me.

BSZ=64
tsize=(48,48)    

clahe = cv2.createCLAHE(clipLimit=0.01, tileGridSize=(8,8))    
def claheImage(img):
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    gray = gray.astype(np.uint16)
    eq = clahe.apply(gray)
    eq = cv2.cvtColor(eq, cv2.COLOR_GRAY2BGR)
    eq = eq.astype(np.float32)
    return eq

val_generator = ImageDataGenerator(rescale=1/255,
                                  preprocessing_function=claheImage)

tr_generator = ImageDataGenerator(rescale = 1/255,
                                  zoom_range=0.3,
                                  shear_range=0.3,
                                  horizontal_flip=True,
                                  rotation_range=15,
                                  fill_mode="nearest",
                                  preprocessing_function=claheImage)

val_data = val_generator.flow_from_directory("./test",
                                             batch_size= BSZ,
                                             target_size=tsize,
                                             color_mode="rgb",
                                             interpolation="nearest")
tr_data = tr_generator.flow_from_directory("./train",
                                          batch_size= BSZ,
                                          target_size=tsize,
                                          color_mode="rgb",
                                          interpolation="nearest")

When I use this code I get an empty image and warning: "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers)"

for _ in range(1):
    img, label = next(val_data)
    print(img.shape)
    plt.imshow(img[0])
    type(img[0])
    plt.show()

But when I use astype('unit8') I can display it. So, Im not sure if CLAHE is being applied correctly.

for _ in range(1):
    img, label = next(val_data)
    print(img.shape)
    #plt.imshow(img[0])
    plt.imshow(img[0].astype('uint8'))
    type(img[0])
    plt.show()

More details can be found in the following link.

https://colab.research.google.com/drive/1RguAZ9_9pREQNDkP6ort2mY1ffsFOLms?usp=sharing

CodePudding user response：

This is because of Matplotlib. The range of pixel values should be between [0, 255] for integers and [0., 1.] for floating point. Your preprocessing function must not respect this.

I suggest you remove rescale=1/255, but make sure the range of values that's output from your function is between 0 and 1.

def claheImage(img):
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    gray = gray.astype(np.uint16)
    eq = clahe.apply(gray)
    eq = cv2.cvtColor(eq, cv2.COLOR_GRAY2BGR)
    eq = eq.astype(np.float32)
    eq = eq / np.max(eq) # this makes sure the output is between 0 and 1
    return eq