Stochastic Interpolants via Conditional Dependent Coupling

2025-09-27

AI Summary

Current image generation models struggle to balance computational efficiency and generation quality: VAEs suffer from information loss and limited end-to-end trainability; pixel-space diffusion models incur high computational overhead; and cascaded architectures face distribution mismatch, knowledge fragmentation, and difficulties in joint optimization due to their staged design. To address these limitations, we propose a unified multi-stage diffusion framework grounded in conditional dependency coupling. Our approach models image generation as a multi-step interpolation trajectory and implements it via a single Diffusion Transformer that enables cross-stage parameter sharing and end-to-end joint optimization. Leveraging stochastic interpolation and conditional coupling, the framework performs multi-scale modeling directly in pixel space. Experiments demonstrate that our method achieves high-fidelity generation across diverse resolutions while maintaining efficient inference—significantly outperforming state-of-the-art VAE- and cascade-based systems.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Existing image generation models face critical challenges regarding the trade-off between computation and fidelity. Specifically, models relying on a pretrained Variational Autoencoder (VAE) suffer from information loss, limited detail, and the inability to support end-to-end training. In contrast, models operating directly in the pixel space incur prohibitive computational cost. Although cascade models can mitigate computational cost, stage-wise separation prevents effective end-to-end optimization, hampers knowledge sharing, and often results in inaccurate distribution learning within each stage. To address these challenges, we introduce a unified multistage generative framework based on our proposed Conditional Dependent Coupling strategy. It decomposes the generative process into interpolant trajectories at multiple stages, ensuring accurate distribution learning while enabling end-to-end optimization. Importantly, the entire process is modeled as a single unified Diffusion Transformer, eliminating the need for disjoint modules and also enabling knowledge sharing. Extensive experiments demonstrate that our method achieves both high fidelity and efficiency across multiple resolutions.

Problem

• Addresses trade-offs between computational cost and image fidelity
• Resolves information loss and limited detail in VAE-based models
• Enables end-to-end optimization in multistage generative frameworks

Innovation

• Multistage generative framework with Conditional Dependent Coupling
• Unified Diffusion Transformer enabling end-to-end optimization
• Interpolant trajectories ensuring accurate distribution learning