// Facebook Research Hand Tracking Challenge Toolkit - evaluation and visualization tools for 3D hand tracking. Supports loading HOT3D data, computing metrics (PA-MPJPE, AUC, etc.), visualizing 3D pose projections, and generating tracking evaluation reports. Essential for benchmarking hand tracking algorithms.
| name | hand-tracking-toolkit |
| description | Facebook Research Hand Tracking Challenge Toolkit - evaluation and visualization tools for 3D hand tracking. Supports loading HOT3D data, computing metrics (PA-MPJPE, AUC, etc.), visualizing 3D pose projections, and generating tracking evaluation reports. Essential for benchmarking hand tracking algorithms. |
| license | Apache 2.0 |
| metadata | {"skill-author":"K-Dense Inc.","original-repo":"https://github.com/facebookresearch/hand_tracking_toolkit","organization":"Meta Facebook Research","skill-category":"Computer Vision / Evaluation","tags":["hand-tracking","evaluation","metrics","visualization","benchmarking","hot3d"]} |
Comprehensive toolkit from Meta Facebook Research for evaluating and visualizing 3D hand tracking systems. Provides standardized metrics, visualization tools, and data loaders for the HOT3D dataset. Essential for researchers developing and benchmarking hand tracking algorithms on multi-view egocentric data.
Use this for: Evaluating hand tracking performance, generating evaluation reports, visualizing 3D predictions vs ground truth.
Use when you need to:
Compute widely-used hand tracking metrics:
Rich visualization options:
Easy data loading:
# Clone repository
git clone https://github.com/facebookresearch/hand_tracking_toolkit.git
cd hand_tracking_toolkit
# Install
pip install -r requirements.txt
# Run evaluation
python evaluate.py \
--predictions path/to/predictions.pkl \
--ground_truth path/to/hot3d/sequence \
--output_dir ./results
# Generate visualizations
python visualize.py \
--predictions path/to/predictions.pkl \
--ground_truth path/to/hot3d/sequence \
--output visualizations.png
from toolkit import Evaluator
import pickle
# Load predictions
with open('predictions.pkl', 'rb') as f:
predictions = pickle.load(f)
# Load ground truth
evaluator = Evaluator()
evaluator.load_ground_truth('path/to/hot3d_sequence')
# Compute metrics
metrics = evaluator.evaluate(predictions)
print(f"PA-MPJPE: {metrics['pa_mpjpe']:.2f} mm")
print(f"AUC: {metrics['auc']:.3f}")
print(f"PCK@0.1: {metrics['pck_01']*100:.1f}%")
from toolkit import Visualizer
viz = Visualizer()
# Load data
viz.load_predictions('predictions.pkl')
viz.load_ground_truth('ground_truth_path')
# Create visualization
fig = viz.plot_3d_skeleton(
frame_id=100,
show_pred=True,
show_gt=True,
show_errors=True
)
fig.savefig('comparison_3d.png')
from toolkit import ReportGenerator
report = ReportGenerator()
report.load_evaluation_results('results.json')
# Generate PDF report
report.generate_pdf(
output_path='evaluation_report.pdf',
include_plots=True,
include_per_joint_errors=True
)
Prediction format:
predictions = {
'sequence_id': 'seq001',
'frames': [
{
'frame_id': 0,
'left_hand': np.array((21, 3)), # 21 joints x 3 coords
'right_hand': np.array((21, 3)),
'confidence': np.array(21),
},
# ... more frames
]
}
| Metric | Description | Lower is Better |
|---|---|---|
| MPJPE | Mean per-joint position error (mm) | ✓ |
| PA-MPJPE | Aligned MPJPE (procrustes) | ✓ |
| AUC | Area under error threshold curve | ✗ |
| PCK | % keypoints within threshold | ✗ |
Works with:
Apache 2.0
Comprehensive citation management for academic research. Search Google Scholar and PubMed for papers, extract accurate metadata, validate citations, and generate properly formatted BibTeX entries. This skill should be used when you need to find papers, verify citation information, convert DOIs to BibTeX, or ensure reference accuracy in scientific writing.
Core skill for the deep research and writing tool. Write scientific manuscripts in full paragraphs (never bullet points). Use two-stage process with (1) section outlines with key points using research-lookup then (2) convert to flowing prose. IMRAD structure, citations (APA/AMA/Vancouver), figures/tables, reporting guidelines (CONSORT/STROBE/PRISMA), for research papers and journal submissions.
Egocentric Hand-Object Segmentation (EgoHOS) - pixel-level hand and object segmentation in egocentric videos. Outputs fine-grained segmentation masks with hand regions highlighted. Specialized for hand-object interaction scenarios with pixel-accurate masks. Ideal for detailed interaction analysis.
High-quality 3D hand pose estimation for egocentric videos from ECCV 2024 (ap229997/hands). Provides 3D joint keypoints and skeleton visualization projected to 2D. Optimized for daily egocentric activities with state-of-the-art accuracy. Outputs hand skeleton overlays on video frames.
Real-time hand detection in egocentric videos using victordibia/handtracking. Outputs bounding boxes for hands, specifically trained on EgoHands dataset. Supports video input/output with labeled hand boxes. Lightweight and fast for egocentric view applications.
HOT3D (Hand-Object 3D Dataset) by Meta Facebook - multi-view egocentric hand and object 3D tracking for Aria/Quest smart glasses. State-of-the-art multi-view 3D hand pose, object pose, and hand-object interaction tracking. Supports visualization with 3D joint projections, meshes, and skeletal overlays on video frames.