DiffFaceSketch: High-Fidelity Face Image Synthesis with Sketch-Guided Latent Diffusion Model


Sketch-Guided Lantent Diffusion Model (SGLDM) synthesizes high-quality face images with high consistency of input sketches.


Synthesizing face images from monochrome sketches is one of the most fundamental tasks in the field of image-to-image translation. However, it is still challenging to (1)~make models learn the high-dimensional face features such as geometry and color, and (2)~take into account the characteristics of input sketches. Existing methods often use sketches as indirect inputs (or as auxiliary inputs) to guide the models, resulting in the loss of sketch features or the alteration of geometry information. In this paper, we introduce a Sketch-Guided Latent Diffusion Model (SGLDM), a LDM-based network architect trained on the paired sketch-face dataset. We apply a Multi-Auto-Encoder (AE) to encode the different input sketches from different regions of a face from pixel space to a feature map in latent space, which enables us to reduce the dimension of the sketch input while preserving the geometry-related information of local face details. We build a sketch-face paired dataset based on the existing method that extracts the edge map from an image. We then introduce a Stochastic Region Abstraction (SRA), an approach to augment our dataset to improve the robustness of SGLDM to handle sketch input with arbitrary abstraction. The evaluation study shows that SGLDM can synthesize high-quality face images with different expressions, facial accessories, and hairstyles from various sketches with different abstraction levels.

The framework of SGLDM. In the sketch embedding stage (a), given a sketch input S, a pretrained sketch encoder ζ encodes S into a feature vector S'. The decoder θ then decoder S' into a feature map S˜. In the latent denoising stage (b), the random latent code ZT is concatenated by the feature map S˜ and denoised to ζ by a U-Net. Finally, the output latent code Z0 is decoded by a decoder D to the final output.

Distribution Mapping

The different feature distribution mapping between jointly trained conditional embedding (dashed line) and the separately trained conditional embedding (solid line), from the sketch (a) to the image (b) domain.

Stochastic Region Abstraction Data Augmentation

The original image in Celeba-HQ and its extracted edge map (a), and the result of paired data after cleaning up the background (b). Sketch simplification results from 3 different resolutions faces (left-bottom). And the random seamed data samples (right-bottom).

Sketch Abstraction Robustness

We compared the synthetic faces of the 2-Stage trained model and the jointly trained model.

Editing Capability

Examples of partial editing such as hairstyles, earings, and expressions.

Comparasion Study

We compared the synthetic faces of the proposed SGLDM with the state-of-the-art methods. From the results, we confirm that SGLDM can synthesizes noiseless faces maintaining maximum consistency with the input sketch, except for some facial details such as nasolabial folds.


            doi = {10.48550/ARXIV.2302.06908},
            url = {https://arxiv.org/abs/2302.06908},
            title = {DiffFaceSketch: High-Fidelity Face Image Synthesis with Sketch-Guided Latent Diffusion Model},
            year = {2023},