Is This Loss Informative? Faster Text-to-Image Customization by Tracking Objective Dynamics
This work addresses a practical bottleneck for researchers and practitioners using text-to-image models by reducing GPU resource usage and experiment time, though it is incremental as it builds on existing personalization methods.
The paper tackles the problem of long training times in text-to-image personalization methods like Textual Inversion and DreamBooth, proposing an early stopping criterion that speeds up adaptation by up to 8 times without significant quality loss, as demonstrated on Stable Diffusion with 48 concepts.
Text-to-image generation models represent the next step of evolution in image synthesis, offering a natural way to achieve flexible yet fine-grained control over the result. One emerging area of research is the fast adaptation of large text-to-image models to smaller datasets or new visual concepts. However, many efficient methods of adaptation have a long training time, which limits their practical applications, slows down experiments, and spends excessive GPU resources. In this work, we study the training dynamics of popular text-to-image personalization methods (such as Textual Inversion or DreamBooth), aiming to speed them up. We observe that most concepts are learned at early stages and do not improve in quality later, but standard training convergence metrics fail to indicate that. Instead, we propose a simple drop-in early stopping criterion that only requires computing the regular training objective on a fixed set of inputs for all training iterations. Our experiments on Stable Diffusion for 48 different concepts and three personalization methods demonstrate the competitive performance of our approach, which makes adaptation up to 8 times faster with no significant drops in quality.