DreamText: High Fidelity Scene Text Synthesis

TL;DR

DreamText tackles high-fidelity scene text synthesis by introducing explicit character-level guidance into diffusion training. It reconstructs the diffusion process to expose and rectify character attention through latent character masks derived from cross-attention, while jointly training a font-diverse text encoder and generator. A suite of losses—masked diffusion, cross-attention, cross-modal alignment, and character identity—paired with a heuristic alternate optimization enables robust learning of character positions and text regions. The approach yields superior qualitative and quantitative performance on polystylistic text datasets, highlighting its potential for reliable text editing and synthesis in complex images.

Abstract

Scene text synthesis involves rendering specified texts onto arbitrary images. Current methods typically formulate this task in an end-to-end manner but lack effective character-level guidance during training. Besides, their text encoders, pre-trained on a single font type, struggle to adapt to the diverse font styles encountered in practical applications. Consequently, these methods suffer from character distortion, repetition, and absence, particularly in polystylistic scenarios. To this end, this paper proposes DreamText for high-fidelity scene text synthesis. Our key idea is to reconstruct the diffusion training process, introducing more refined guidance tailored to this task, to expose and rectify the model's attention at the character level and strengthen its learning of text regions. This transformation poses a hybrid optimization challenge, involving both discrete and continuous variables. To effectively tackle this challenge, we employ a heuristic alternate optimization strategy. Meanwhile, we jointly train the text encoder and generator to comprehensively learn and utilize the diverse font present in the training dataset. This joint training is seamlessly integrated into the alternate optimization process, fostering a synergistic relationship between learning character embedding and re-estimating character attention. Specifically, in each step, we first encode potential character-generated position information from cross-attention maps into latent character masks. These masks are then utilized to update the representation of specific characters in the current step, which, in turn, enables the generator to correct the character's attention in the subsequent steps. Both qualitative and quantitative results demonstrate the superiority of our method to the state of the art.

Figures (14)

• Figure 1: Displayed are the results generated using our DreamText, showcasing its prowess across varied inputs.
• Figure 2: Current methods encounter significant challenges, i.e., character repetition and absence (top) and character distortion (bottom).
• Figure 3: The problematic results rendered by characters' attention maps (AMs).
• Figure 4: The mIoU scores of UDiffText, TextDiffuser, and our method on LAION-OCR and SynthText over global training steps. Our method adopts a balanced supervision strategy: we initially use latent character masks to steer the character's attention for a warm-up in the earlier stage and stop guiding after 25,000 steps.
• Figure 5: An overview of proposed heuristic alternate optimization strategy.