DiffSketch: Representative Feature Extraction During Diffusion Process for Sketch Extraction with One Example

Kwan yun, Youngseo Kim, Kwanggyoon Seo, Chang Wook Seo, Junyoung Noh*
Visual Media Lab, KAIST
Results of DiffSketch and distilled DiffSketch, trained with one example.

Abstract

We introduce DiffSketch, a method for generating a variety of stylized sketches from images. Our approach focuses on selecting representative features from the rich semantics of deep features within a pretrained diffusion model. This novel sketch generation method can be trained with one manual drawing. Furthermore, efficient sketch extraction is ensured by distilling a trained generator into a streamlined extractor. We select denoising diffusion features through analysis and integrate these selected features with VAE features to produce sketches. Additionally, we propose a sampling scheme for training models using a conditional generative approach. Through a series of comparisons, we verify that distilled DiffSketch not only outperforms existing state-of-the-art sketch extraction methods but also surpasses diffusion-based stylization methods in the task of extracting sketches.

Overview of Diffsketch

Overview of DiffSketch

The UNet features generated during the denoising process are fed to the Aggregation networks to be fused with the VAE features to generate a sketch corresponding to the image that Stable Diffusion generates.

Anaylsis on sampled features

wave

PCA is applied on DDIM sampled features from different classes. (a): features colored with human-labeled classes. (b): features colored with denoising timesteps.

wave

Up: features colored with human-labeled classes. Down: features colored with denoising timesteps.

Qualitative comparison with alternative sketch extraction methods

Comparsion1


Qualitative comparison with alternative sketch extraction methods on COCO dataset

Comparsion1


Qualitative comparison with alternative sketch extraction methods on BSDS500 dataset

Comparsion1

More results

video1
video2
video3

BibTeX

@article{aaa
  title={DiffSketch: Representative Feature Extraction During Diffusion Process for Sketch Extraction with One Example},
  author={Anonymous},
  year={2024}
}