Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FaceID-6M: A Large-Scale, Open-Source FaceID Customization Dataset

Shuhe Wang, Xiaoya Li, Jiwei Li, Guoyin Wang, Xiaofei Sun, Bob Zhu, Han Qiu, Mo Yu, Shengjie Shen, Tianwei Zhang, Eduard Hovy

TL;DR

FaceID-6M addresses the lack of public datasets for FaceID customization by releasing a 6-million sample open dataset filtered from LAION-5B. The authors detail a rigorous filtering pipeline for both images and descriptions to ensure high facial fidelity and meaningful text descriptions, and demonstrate effective use in diffusion-based customization via IP-Adapter. Experimental results show models trained on FaceID-6M are comparable to or slightly better than industrial baselines in FaceID fidelity and follow-up human evaluations, with clear scaling benefits as data size increases. By providing code, data, and models publicly, the work aims to enhance transparency and accelerate research in personalized facial generation and FaceID customization.

Abstract

Due to the data-driven nature of current face identity (FaceID) customization methods, all state-of-the-art models rely on large-scale datasets containing millions of high-quality text-image pairs for training. However, none of these datasets are publicly available, which restricts transparency and hinders further advancements in the field. To address this issue, in this paper, we collect and release FaceID-6M, the first large-scale, open-source FaceID dataset containing 6 million high-quality text-image pairs. Filtered from LAION-5B \cite{schuhmann2022laion}, FaceID-6M undergoes a rigorous image and text filtering steps to ensure dataset quality, including resolution filtering to maintain high-quality images and faces, face filtering to remove images that lack human faces, and keyword-based strategy to retain descriptions containing human-related terms (e.g., nationality, professions and names). Through these cleaning processes, FaceID-6M provides a high-quality dataset optimized for training powerful FaceID customization models, facilitating advancements in the field by offering an open resource for research and development. We conduct extensive experiments to show the effectiveness of our FaceID-6M, demonstrating that models trained on our FaceID-6M dataset achieve performance that is comparable to, and slightly better than currently available industrial models. Additionally, to support and advance research in the FaceID customization community, we make our code, datasets, and models fully publicly available. Our codes, models, and datasets are available at: https://github.com/ShuheSH/FaceID-6M.

FaceID-6M: A Large-Scale, Open-Source FaceID Customization Dataset

TL;DR

FaceID-6M addresses the lack of public datasets for FaceID customization by releasing a 6-million sample open dataset filtered from LAION-5B. The authors detail a rigorous filtering pipeline for both images and descriptions to ensure high facial fidelity and meaningful text descriptions, and demonstrate effective use in diffusion-based customization via IP-Adapter. Experimental results show models trained on FaceID-6M are comparable to or slightly better than industrial baselines in FaceID fidelity and follow-up human evaluations, with clear scaling benefits as data size increases. By providing code, data, and models publicly, the work aims to enhance transparency and accelerate research in personalized facial generation and FaceID customization.

Abstract

Due to the data-driven nature of current face identity (FaceID) customization methods, all state-of-the-art models rely on large-scale datasets containing millions of high-quality text-image pairs for training. However, none of these datasets are publicly available, which restricts transparency and hinders further advancements in the field. To address this issue, in this paper, we collect and release FaceID-6M, the first large-scale, open-source FaceID dataset containing 6 million high-quality text-image pairs. Filtered from LAION-5B \cite{schuhmann2022laion}, FaceID-6M undergoes a rigorous image and text filtering steps to ensure dataset quality, including resolution filtering to maintain high-quality images and faces, face filtering to remove images that lack human faces, and keyword-based strategy to retain descriptions containing human-related terms (e.g., nationality, professions and names). Through these cleaning processes, FaceID-6M provides a high-quality dataset optimized for training powerful FaceID customization models, facilitating advancements in the field by offering an open resource for research and development. We conduct extensive experiments to show the effectiveness of our FaceID-6M, demonstrating that models trained on our FaceID-6M dataset achieve performance that is comparable to, and slightly better than currently available industrial models. Additionally, to support and advance research in the FaceID customization community, we make our code, datasets, and models fully publicly available. Our codes, models, and datasets are available at: https://github.com/ShuheSH/FaceID-6M.

Paper Structure

This paper contains 24 sections, 8 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Text-to-image Generation
  4. FaceID Customization
  5. Dataset Construction: FaceID-6M
  6. Text-Image Pairs Collecting
  7. Language Filtering
  8. Image Filtering
  9. (1) Face Detection.
  10. (2) Minimum Face Size Constraints.
  11. (3) Resolution Constraints.
  12. Text Filtering
  13. FaceID Customization based on FaceID-6M
  14. Preliminaries: Diffusion Models
  15. FaceID-6M Based FaceID Customization
  16. ...and 9 more sections

Figures (4)

  • Figure 1: Statistics for the English subset of LAION-5B, presenting (1) image width, (2) image height, and (3) text length from left to right.
  • Figure 2: Images sampled from our constructed FaceID-6M dataset are presented.
  • Figure 3: The results demonstrate the performance of FaceID customization models in maintaining FaceID fidelity. For models, "InstantID" refers to the official InstantID model, while "InstantID + FaceID-6M" represents the model further fine-tuned on our FaceID-6M dataset. These results indicate that the model trained on our constructed FaceID-6M dataset achieves comparable performance to the official InstantID model in preserving FaceID fidelity.
  • Figure 4: Scaling results by sampling subsets of different sizes from FaceID-6M: (1) 1K, (2) 10K, (3) 100K, (4) 1M, (5) 2M, (6) 4M, and (7) the full dataset (6M).