Face Pointing

Overview

The Head Pose Database consists of 15 sets of images, with each set containing 2 series of 93 images of the same person at different head poses. The database is used for head pose estimation and face orientation detection research.

Total Images: 2,790 images across 15 persons
Format: JPEG images
Size: Approximately 30 MB
Annotations: Each image filename encodes person ID, tilt angle, and pan angle

Data Structure

When accessing an example using ds[i], you will receive a dictionary with the following keys:

Key	Type	Description
`image`	`PIL.Image.Image`	RGB image of person’s head at specific pose
`person_id`	int	Person identifier (1-15)
`angles`	List[int]	[tilt_angle, pan_angle] representing head orientation

Usage Example

Basic Usage

from stable_datasets.images.face_pointing import FacePointing

# Load the training split
ds = FacePointing(split="train")

# Access a single sample
sample = ds[0]
image = sample["image"]  # PIL Image
person_id = sample["person_id"]  # 1-15
angles = sample["angles"]  # [tilt, pan]

print(f"Dataset size: {len(ds)}")
print(f"Person {person_id}, Angles: {angles}")

With PyTorch

from stable_datasets.images.face_pointing import FacePointing
import torch
from torch.utils.data import DataLoader

# Load dataset and format for PyTorch
ds = FacePointing(split="train").with_format("torch")

# Create a DataLoader
dataloader = DataLoader(ds, batch_size=32, shuffle=True)

for batch in dataloader:
    images = batch["image"]
    person_ids = batch["person_id"]
    angles = batch["angles"]
    # Your training loop here
    break

Analyzing Head Poses

from stable_datasets.images.face_pointing import FacePointing
import numpy as np

ds = FacePointing(split="train")

# Collect angle statistics
tilt_angles = []
pan_angles = []

for sample in ds:
    tilt, pan = sample["angles"]
    tilt_angles.append(tilt)
    pan_angles.append(pan)

print(f"Tilt range: [{min(tilt_angles)}, {max(tilt_angles)}]")
print(f"Pan range: [{min(pan_angles)}, {max(pan_angles)}]")

Dataset Details

Source: Head Pose Image Database
Homepage: http://crowley-coutaz.fr/HeadPoseDataSet/
Image Format: JPEG, RGB color
Total Images: 2,790 images
Persons: 15 different individuals
Series per Person: 2 series of 93 images each
Organization: Files organized in directories by person
Special: Includes a “Front” directory with 30 frontal images (pan and tilt angles equal to 0)

Angle Information

Each image filename encodes the head pose angles:

Tilt angle: Vertical head rotation (up/down)
Pan angle: Horizontal head rotation (left/right)

Angles are encoded in the filename with the pattern: personXXYYY[+-]ZZ[+-]WW.jpg where:

XX: Person ID (01-15)
YYY: Image number within series
[+-]ZZ: Tilt angle in degrees
[+-]WW: Pan angle in degrees

Citation

@inproceedings{gourier2004estimating,
    title={Estimating face orientation from robust detection of salient facial features},
    author={Gourier, Nicolas and Hall, Daniela and Crowley, James L},
    booktitle={ICPR International Workshop on Visual Observation of Deictic Gestures},
    year={2004},
    organization={Citeseer}
}