Generative Video Bi-flow

2025-03-09

AI Summary

To address frame-wise drift accumulation and high computational overhead in video generation, this paper proposes the first unconditional streaming video generation framework based on a dual-stream neural ordinary differential equation (ODE). Methodologically, it introduces coupled dual-path ODE dynamics: a primary path directly models temporal evolution, while an auxiliary path jointly learns residual corrections; dynamic error suppression is achieved via noise injection and a bilinear optimization objective. This design eliminates reliance on multi-step noise sampling inherent in conventional diffusion models, substantially reducing the number of ODE solver steps. Experiments demonstrate that our method generates high-fidelity videos significantly faster than prior approaches—achieving notable inference acceleration—across multiple benchmark datasets. The visual quality matches state-of-the-art conditional diffusion models, marking the first successful realization of an efficient, robust, and streaming-capable ODE-based paradigm for video generation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Natural Language Processing: Generation

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Data management and stream processing for Web, mobile and wireless applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We propose a novel generative video model by robustly learning temporal change as a neural Ordinary Differential Equation (ODE) flow with a bilinear objective of combining two aspects: The first is to map from the past into future video frames directly. Previous work has mapped the noise to new frames, a more computationally expensive process. Unfortunately, starting from the previous frame, instead of noise, is more prone to drifting errors. Hence, second, we additionally learn how to remove the accumulated errors as the joint objective by adding noise during training. We demonstrate unconditional video generation in a streaming manner for various video datasets, all at competitive quality compared to a baseline conditional diffusion but with higher speed, i.e., fewer ODE solver steps.

Problem

• Learning temporal change in videos using neural ODE flow.
• Reducing drifting errors by starting from past frames.
• Enhancing video generation speed with fewer ODE steps.

Innovation

• Neural ODE flow for temporal change learning
• Bilinear objective combining past-to-future mapping
• Error removal via noise addition during training