prithivMLmods/Face-3D-Unified-Preferences

Face-3D-Unified-Preferences

Face-3D-Unified-Preferences is a high-quality dataset designed for human face depth estimation, 3D face reconstruction, and unified preference learning. The dataset is a mixture of male and female human face portraits, providing a diverse collection of facial appearances, identities, poses, and expressions for training modern computer vision and multimodal AI models. Each sample contains an RGB face image, a dense facial depth map, and a corresponding 3D mesh in GLB format, enabling direct supervision for geometry-aware learning and image-to-3D reconstruction tasks. The unified nature of the dataset makes it well suited for developing robust models that generalize across different facial characteristics while supporting research in depth estimation, facial geometry prediction, neural rendering, synthetic data generation, and multimodal understanding. The dataset is distributed in the Hugging Face Datasets format using optimized Parquet files for efficient loading, while the associated GLB meshes can be downloaded individually when required.

Dataset Statistics

Property	Value
Number of Samples	5,000
Image Format	RGB
3D Mesh Format	GLB
Dataset Format	Optimized Parquet

Dataset Structure

Each sample contains the following fields:

Column	Type	Description
image	Image	Original RGB human face image
output_mask	Image	Dense facial depth map
glb_3d_file	String	Filename of the corresponding GLB mesh

Example:

sample = ds[0]

print(sample.keys())

# dict_keys([
#     "image",
#     "output_mask",
#     "glb_3d_file"
# ])

Loading the Dataset

from datasets import load_dataset

ds = load_dataset(
    "prithivMLmods/Face-3D-Unified-Preferences",
    split="train"
)

Example Usage

from datasets import load_dataset
from huggingface_hub import hf_hub_download
import matplotlib.pyplot as plt

ds = load_dataset(
    "prithivMLmods/Face-3D-Unified-Preferences",
    split="train"
)

sample = ds[0]

image = sample["image"]
mask = sample["output_mask"]
glb_name = sample["glb_3d_file"]

print("GLB filename:", glb_name)

fig, axes = plt.subplots(1, 2, figsize=(10, 5))

axes[0].imshow(image)
axes[0].set_title("Image")
axes[0].axis("off")

axes[1].imshow(mask)
axes[1].set_title("Output Depth Map")
axes[1].axis("off")

plt.show()

glb_path = hf_hub_download(
    repo_id="prithivMLmods/Face-3D-Unified-Preferences",
    filename=glb_name,
    repo_type="dataset",
)

print("Downloaded mesh to:", glb_path)

Downloading the 3D Mesh

The glb_3d_file field contains the filename of the corresponding 3D face reconstruction. The mesh can be downloaded directly from the dataset repository using hf_hub_download.

from huggingface_hub import hf_hub_download

mesh_path = hf_hub_download(
    repo_id="prithivMLmods/Face-3D-Unified-Preferences",
    filename=sample["glb_3d_file"],
    repo_type="dataset",
)

print(mesh_path)

The downloaded GLB file can be viewed or processed using Blender, MeshLab, Three.js, Unity, Unreal Engine, or any software that supports the GLB format.

Dataset Features

• High-quality human face RGB images
• Mixture of male and female face portraits
• Dense facial depth maps
• One-to-one correspondence between RGB images and depth maps
• A corresponding GLB mesh for every sample
• Optimized Parquet dataset format for efficient loading
• Compatible with the Hugging Face Datasets library
• Suitable for image-to-3D reconstruction, face depth estimation, and unified preference learning
• Ready for large-scale computer vision and multimodal AI pipelines

License

This dataset is released under the Apache-2.0 License.