Title: Diverse Diffusion: Enhancing Image Diversity in Text-to-Image Generation
Abstract: Generative modeling methods can generate images from textual or visual
inputs. However, diversity in the generated images persists as a major challenge of the existing approaches. We address this issue head-on and demonstrating that
-
the diversity of a generated batch of images is intrinsically linked to the diversity within the latent variables
-
leveraging the geometry of the latent space, we can establish an effective metric for quantifying diversity; and
-
employing this insight allows one to achieve a significantly enhanced diversity in image generation beyond the capabilities of traditional random independent sampling.
This advancement is consistent across a variety of generative models, including latent diffusion models and GANs.
Additionally, we have integrated our contributions into a widely recognized tool for generative image modeling, ensuring that our improvements are accessible to the broader community. As a result, this work not only presents a methodological advancement in
generative modeling but also significantly broadens the scope of potential applications by enhancing the diversity of generated images
Short
bio: Mariia Zameshina recently completed her PhD at University Gustave Eiffel and Meta (Facebook AI Research). During her PhD, her main research focus was on ethical AI, including improving the fairness and diversity
of generative models and preserving privacy using these models. Before that, she completed her master's degree at Grenoble INP, where her thesis was on explainable learning. She has also completed internships in computer vision at Google and Align Technology.
________________________________________________________________________________
ID de réunion :
342 491 866 837
Code secret :
E8CxW4
ID de la conférence téléphonique :
742 628 992#
________________________________________________________________________________