Search papers, labs, and topics across Lattice.
The paper introduces Laplacian Multi-scale Flow Matching (LapFlow), a novel flow matching framework for image generation that leverages Laplacian pyramid decomposition to process image scales in parallel using a mixture-of-transformers architecture with causal attention. This parallel processing eliminates the need for explicit renoising between scales, improving generation quality and accelerating sampling compared to cascaded approaches. Experiments on CelebA-HQ and ImageNet demonstrate that LapFlow achieves superior sample quality with fewer GFLOPs and faster inference, scaling effectively to high-resolution image generation.
Ditch the renoising bottleneck: LapFlow's parallel multi-scale architecture slashes GFLOPs and speeds up inference in flow matching for high-resolution image generation.
In this paper, we present Laplacian multiscale flow matching (LapFlow), a novel framework that enhances flow matching by leveraging multi-scale representations for image generative modeling. Our approach decomposes images into Laplacian pyramid residuals and processes different scales in parallel through a mixture-of-transformers (MoT) architecture with causal attention mechanisms. Unlike previous cascaded approaches that require explicit renoising between scales, our model generates multi-scale representations in parallel, eliminating the need for bridging processes. The proposed multi-scale architecture not only improves generation quality but also accelerates the sampling process and promotes scaling flow matching methods. Through extensive experimentation on CelebA-HQ and ImageNet, we demonstrate that our method achieves superior sample quality with fewer GFLOPs and faster inference compared to single-scale and multi-scale flow matching baselines. The proposed model scales effectively to high-resolution generation (up to 1024$\times$1024) while maintaining lower computational overhead.
