L-CoDe:Language-Based Colorization Using Color-Object Decoupled Conditions

Colorizing a grayscale image is inherently an ill-posed problem with multi-modal uncertainty. Language-based colorization offers natural way of interaction to reduce such uncertainty via user-provided caption. However, the color-object coupling and mismatch issues make mapping from word color difficult. In this paper, we propose L-CoDe, Colorization network using Decoupled conditions. A predictor for object-color corresponding matrix (OCCM) novel attention transfer module (ATM) are introduced solve problem. To deal that results in incorrect correspondence, adopt soft-gated injection (SIM). We further present new dataset containing annotated pairs provide supervisory signals resolving Experimental show our approach outperforms state-of-the-art methods conditioned on captions.