Toward Spatially Unbiased Generative Models
This addresses a specific issue in generative models for computer vision, offering an incremental improvement by modifying existing architectures to reduce spatial bias.
The paper tackles the problem of spatial bias in image generation models, where generators perform poorly at unseen locations and scales due to translation-variant implicit positional encodings, and proposes injecting explicit positional encoding to achieve spatially unbiased generation, enabling robust performance in tasks like GAN inversion and multi-scale generation.
Recent image generation models show remarkable generation performance. However, they mirror strong location preference in datasets, which we call spatial bias. Therefore, generators render poor samples at unseen locations and scales. We argue that the generators rely on their implicit positional encoding to render spatial content. From our observations, the generator's implicit positional encoding is translation-variant, making the generator spatially biased. To address this issue, we propose injecting explicit positional encoding at each scale of the generator. By learning the spatially unbiased generator, we facilitate the robust use of generators in multiple tasks, such as GAN inversion, multi-scale generation, generation of arbitrary sizes and aspect ratios. Furthermore, we show that our method can also be applied to denoising diffusion probabilistic models.