Condition-Aware Neural Network for Controlled Image Generation

TL;DR

This paper introduces Condition-Aware Neural Network (CAN), a novel method for controlling image generation by dynamically adjusting neural network weights based on input conditions, improving efficiency and quality in diffusion transformer models.

Contribution

The paper proposes a new condition-aware weight generation module that enhances control in image generative models, achieving better quality with fewer computations.

Findings

CAN improves image generation quality on ImageNet and COCO datasets.

CAN with EfficientViT achieves 2.78 FID on ImageNet 512x512, surpassing larger models.

Significantly reduces computational cost by 52x per sampling step.

Abstract

We present Condition-Aware Neural Network (CAN), a new method for adding control to image generative models. In parallel to prior conditional control methods, CAN controls the image generation process by dynamically manipulating the weight of the neural network. This is achieved by introducing a condition-aware weight generation module that generates conditional weight for convolution/linear layers based on the input condition. We test CAN on class-conditional image generation on ImageNet and text-to-image generation on COCO. CAN consistently delivers significant improvements for diffusion transformer models, including DiT and UViT. In particular, CAN combined with EfficientViT (CaT) achieves 2.78 FID on ImageNet 512x512, surpassing DiT-XL/2 while requiring 52x fewer MACs per sampling step.