Abstract
The problem of colorizing monochrome images attracts significant attention
and holds great potential for applications due to over a century of black-and-white
photography. However, because its inherently ill-posed nature, existing solutions
often lack consistency and realism. Furthermore, the opportunity for user guidance
is frequently overlooked, despite its potential to simplify the problem and enhance
user satisfaction. In this work, a new solution for colorization with guidance from
colorful dots (hints) is presented. The developed model produces the best qualitative
results, accurately follows the hints, and generates natural-looking images based
on user feedback. To evaluate existing hint-based models, a new metric called Hint
Gain was developed, which numerically assesses the models’ ability to follow and
benefit from user guidance. Finally, a psychophysical experiment was conducted
that verified the capabilities of the developed model and the potential of the metric.
