Synthetic Image Dataset Generation for deep learning applications

This paper presents a novel approach to dataset generation using Denoising Diffusion Probabilistic Models (DDPM) augmented with a tree-branching mechanism. This architecture addresses key challenges in deep learning, such as dataset bias, variance, and scalability, by generating diverse, balanced, and customizable datasets directly from textual inputs. The branching mechanism enables the progressive creation of multiple image variants, ensuring meaningful variability while maintaining consistency with the input prompt. By leveraging different noise schedules and iterative denoising, the proposed model achieves efficient image dataset generation. Initial experiments validate the potential of this architecture to produce datasets suitable for robust model training. This work not only extends the capabilities of DDPM for dataset synthesis but also provides a scalable, domain-agnostic framework for meeting the increasing demands of modern deep learning applications.