LightGen: Efficient Image Generation through Knowledge Distillation and Direct Preference Optimization

• URL: http://arxiv.org/abs/2503.08619v1
• Date: Tue, 11 Mar 2025 16:58:02 GMT
• Title: LightGen: Efficient Image Generation through Knowledge Distillation and Direct Preference Optimization
• Authors: Xianfeng Wu, Yajing Bai, Haoze Zheng, Harold Haodong Chen, Yexin Liu, Zihao Wang, Xuran Ma, Wen-Jie Shu, Xianzu Wu, Harry Yang, Ser-Nam Lim,
• Abstract summary: LightGen is an efficient training paradigm for image generation models.<n>It distills knowledge from state-of-the-art (SOTA) text-to-image models into a compact Masked Autoregressive architecture.<n>Experiments confirm that LightGen achieves image generation quality comparable to SOTA models.
• Score: 37.236005953016175
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Recent advances in text-to-image generation have primarily relied on extensive datasets and parameter-heavy architectures. These requirements severely limit accessibility for researchers and practitioners who lack substantial computational resources. In this paper, we introduce \model, an efficient training paradigm for image generation models that uses knowledge distillation (KD) and Direct Preference Optimization (DPO). Drawing inspiration from the success of data KD techniques widely adopted in Multi-Modal Large Language Models (MLLMs), LightGen distills knowledge from state-of-the-art (SOTA) text-to-image models into a compact Masked Autoregressive (MAR) architecture with only $0.7B$ parameters. Using a compact synthetic dataset of just $2M$ high-quality images generated from varied captions, we demonstrate that data diversity significantly outweighs data volume in determining model performance. This strategy dramatically reduces computational demands and reduces pre-training time from potentially thousands of GPU-days to merely 88 GPU-days. Furthermore, to address the inherent shortcomings of synthetic data, particularly poor high-frequency details and spatial inaccuracies, we integrate the DPO technique that refines image fidelity and positional accuracy. Comprehensive experiments confirm that LightGen achieves image generation quality comparable to SOTA models while significantly reducing computational resources and expanding accessibility for resource-constrained environments. Code is available at https://github.com/XianfengWu01/LightGen

Related papers

• Large-Scale Data-Free Knowledge Distillation for ImageNet via Multi-Resolution Data Generation [53.95204595640208]
  Data-Free Knowledge Distillation (DFKD) is an advanced technique that enables knowledge transfer from a teacher model to a student model without relying on original training data. Previous approaches have generated synthetic images at high resolutions without leveraging information from real images. MUSE generates images at lower resolutions while using Class Activation Maps (CAMs) to ensure that the generated images retain critical, class-specific features.
  arXiv  Detail & Related papers  (2024-11-26T02:23:31Z)
• Analysis of Classifier Training on Synthetic Data for Cross-Domain Datasets [4.696575161583618]
  This study focuses on camera-based traffic sign recognition applications for advanced driver assistance systems and autonomous driving. The proposed augmentation pipeline of synthetic datasets includes novel augmentation processes such as structured shadows and gaussian specular highlights. Experiments showed that a synthetic image-based approach outperforms in most cases real image-based training when applied to cross-domain test datasets.
  arXiv  Detail & Related papers  (2024-10-30T07:11:41Z)
• Diffusion Curriculum: Synthetic-to-Real Generative Curriculum Learning via Image-Guided Diffusion [19.54496184675988]
  Low-quality or scarce data has posed significant challenges for training deep neural networks in practice. Diffusion Curriculum (DisCL) adjusts the image guidance level of image synthesis for each training stage. DisCL focuses on lower-guidance images of high-quality to learn features as a warm-up of learning higher-guidance images that might be weak on diversity or quality.
  arXiv  Detail & Related papers  (2024-10-17T15:33:35Z)
• Meissonic: Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesis [62.06970466554273]
  We present Meissonic, which elevates non-autoregressive masked image modeling (MIM) text-to-image to a level comparable with state-of-the-art diffusion models like SDXL.<n>We leverage high-quality training data, integrate micro-conditions informed by human preference scores, and employ feature compression layers.<n>Our model not only matches but often exceeds the performance of existing models like SDXL in generating high-quality, high-resolution images.
  arXiv  Detail & Related papers  (2024-10-10T17:59:17Z)
• Image-GS: Content-Adaptive Image Representation via 2D Gaussians [55.15950594752051]
  We propose Image-GS, a content-adaptive image representation. Using anisotropic 2D Gaussians as the basis, Image-GS shows high memory efficiency, supports fast random access, and offers a natural level of detail stack. General efficiency and fidelity of Image-GS are validated against several recent neural image representations and industry-standard texture compressors. We hope this research offers insights for developing new applications that require adaptive quality and resource control, such as machine perception, asset streaming, and content generation.
  arXiv  Detail & Related papers  (2024-07-02T00:45:21Z)
• Image Captions are Natural Prompts for Text-to-Image Models [70.30915140413383]
  We analyze the relationship between the training effect of synthetic data and the synthetic data distribution induced by prompts. We propose a simple yet effective method that prompts text-to-image generative models to synthesize more informative and diverse training data. Our method significantly improves the performance of models trained on synthetic training data.
  arXiv  Detail & Related papers  (2023-07-17T14:38:11Z)
• High-Resolution Image Synthesis with Latent Diffusion Models [14.786952412297808]
  Training diffusion models on autoencoders allows for the first time to reach a near-optimal point between complexity reduction and detail preservation. Our latent diffusion models (LDMs) achieve a new state of the art for image inpainting and highly competitive performance on various tasks.
  arXiv  Detail & Related papers  (2021-12-20T18:55:25Z)
• InvGAN: Invertible GANs [88.58338626299837]
  InvGAN, short for Invertible GAN, successfully embeds real images to the latent space of a high quality generative model. This allows us to perform image inpainting, merging, and online data augmentation.
  arXiv  Detail & Related papers  (2021-12-08T21:39:00Z)
• Learning Deformable Image Registration from Optimization: Perspective, Modules, Bilevel Training and Beyond [62.730497582218284]
  We develop a new deep learning based framework to optimize a diffeomorphic model via multi-scale propagation. We conduct two groups of image registration experiments on 3D volume datasets including image-to-atlas registration on brain MRI data and image-to-image registration on liver CT data.
  arXiv  Detail & Related papers  (2020-04-30T03:23:45Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.