Image Colorization Quiz

Answers and explanations appear as you take the quiz.

1. Why is image colorization considered an 'ill-posed' problem?
   • A single grayscale value can correspond to many different real colors
   • Grayscale images contain no usable information at all
   • Color images are too large to process
   • Cameras cannot record grayscale anymore
2. In the popular Lab color space approach, what does the network receive as input?
   • The full RGB image
   • Only the L (lightness) channel
   • Only the red channel
   • A depth map of the scene
3. Why did Zhang et al. frame color prediction as classification over color 'bins' rather than direct regression?
   • Because classification is the only thing neural networks can do
   • To make the images smaller
   • To avoid averaging valid colors into dull, desaturated results
   • To remove the need for training data
4. How do models learn what colors objects should be without being told?
   • They are manually programmed with a list of every object's color
   • They guess randomly each time
   • They read color metadata stored in grayscale files
   • They learn statistical priors from millions of color photos converted to grayscale
5. What is a key challenge when colorizing video rather than single photos?
   • Keeping colors consistent across frames to avoid flickering
   • Videos have no grayscale version
   • Color is impossible to add to moving images
   • Each frame must be a different color
6. What does exemplar-based or hint-based colorization let a user do?
   • Convert color images back to grayscale only
   • Guide the output using reference images or clicked color hints
   • Disable the neural network entirely
   • Translate the image into another language
7. In the CIE Lab setup described in the guide, which channels does the network predict?
   • The L lightness channel only
   • RGB red and green but not blue
   • The a and b chrominance channels
   • The image’s camera pose
8. Why can direct regression toward an average color produce dull colorization?
   • The model is forced to output only black
   • Regression removes the lightness channel
   • The network cannot use training examples
   • Several plausible colors for a gray pixel may average to a muted, less likely color