Facial Wrinkle Segmentation for Cosmetic Dermatology: Pretraining with Texture Map-Based Weak Supervision

TL;DR

This work builds and releases the first public facial wrinkle dataset, 'FFHQ-Wrinkle', an extension of the NVIDIA FFHQ dataset, and introduces a simple training strategy utilizing texture maps, applicable to various segmentation models, to detect wrinkles across the face.

Abstract

Facial wrinkle detection plays a crucial role in cosmetic dermatology. Precise manual segmentation of facial wrinkles is challenging and time-consuming, with inherent subjectivity leading to inconsistent results among graders. To address this issue, we propose two solutions. First, we build and release the first public facial wrinkle dataset, 'FFHQ-Wrinkle', an extension of the NVIDIA FFHQ dataset. It includes 1,000 images with human labels and 50,000 images with automatically generated weak labels. This dataset could serve as a foundation for the research community to develop advanced wrinkle detection algorithms. Second, we introduce a simple training strategy utilizing texture maps, applicable to various segmentation models, to detect wrinkles across the face. Our two-stage training strategy first pretrain models on a large dataset with weak labels (N=50k), or masked texture maps generated through computer vision techniques, without human intervention. We then finetune the models using human-labeled data (N=1k), which consists of manually labeled wrinkle masks. The network takes as input a combination of RGB and masked texture map of the image, comprising four channels, in finetuning. We effectively combine labels from multiple annotators to minimize subjectivity in manual labeling. Our strategies demonstrate improved segmentation performance in facial wrinkle segmentation both quantitatively and visually compared to existing pretraining methods. The dataset is available at https://github.com/labhai/ffhq-wrinkle-dataset.

Figures (7)

• Figure 1: Two-stage training for facial wrinkle segmentation. (a) Weakly supervised pretraining stage: the model learns to extract masked texture maps from RGB face images. (b) Supervised finetuning stage: the model refines its ability to extract facial wrinkles from RGB-masked face images and masked texture maps. The model parameters are initialized with the weights from the weakly supervised pretraining stage.
• Figure 2: Training Dataset. (a) High-resolution face images. (b) Masked texture maps extracted from face images, which include information about facial features. (c) Reliable manual wrinkle masks created by combining the results of multiple annotators.
• Figure 3: Ambiguity in wrinkle evaluation. The labeling results from three annotators for the same image are different.
• Figure 4: Weakly labeled wrinkle generation pipeline. After extracting the texture map from the face image, we mask the non-facial regions to generate a masked texture map containing information on facial features. This masked texture map is then used as a weakly labeled wrinkle.
• Figure 5: Ground truth wrinkle generation pipeline. We combine data labeled by multiple annotators through majority voting to create a reliable ground truth wrinkle.