Practical overview
When it fits
Use it when you want a documented reference implementation for facial landmark localization, a webcam tracking demo, or dataset-based face-alignment evaluation and can work with its specified Python 2.7-era dependencies.
Who it is for
- Computer-vision developers exploring facial landmark detection or webcam-based face tracking.
- Researchers reproducing facial-alignment evaluations on the datasets named in the README.
Documented uses
- Locate facial landmarks in images using a pretrained model.
- Explore face tracking with a local webcam through CameraDemo.py.
- Prepare named facial-landmark datasets and calculate evaluation metrics with the provided scripts.
- Inspect or configure evaluation displays, normalization and failure-threshold settings.
Requirements & limits
The repository code is MIT-licensed. The README says the linked pretrained models were trained on the 300-W dataset, which permits research use only; review the dataset terms before commercial use.
- The README specifies Python 2.7, Theano 0.9.0, Lasagne 0.2 and OpenCV 3.1.0 or newer.
- At least one pretrained model must be downloaded before using the demo or evaluation scripts.
- The evaluation workflow requires separately obtaining and arranging the named datasets.
- The code license and the linked model training dataset have different stated terms: MIT for code and research-only use for the 300-W dataset.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare the environment and model files
Set up the documented prerequisites and obtain a pretrained model before using the demo or evaluation scripts.
Before you start
- Python 2.7
- Theano 0.9.0
- Lasagne 0.2
- OpenCV 3.1.0 or newer
- At least one pretrained model file
Steps
Install Python 2.7; the README recommends Anaconda as a starting point.
Install Theano 0.9.0 and Lasagne using the commands in the repository README, and ensure OpenCV 3.1.0 or newer is available.
Download at least one of the pretrained models linked in the README.
Expected outcome
The documented runtime dependencies and model files are prepared for the repository's demo or experiments.
Guide limitations
- This guide covers preparation only; it does not verify a successful installation or model run.
- The README states that linked models were trained on 300-W, whose terms allow research use only.
Prepare datasets for the documented landmark evaluation
Organize the listed facial-landmark datasets and run the preparation and evaluation scripts described in the README.
Before you start
- Python 2.7 with Theano 0.9.0, Lasagne 0.2 and OpenCV 3.1.0 or newer
- At least one pretrained model file
- The 300W, LFPW, HELEN, AFW and IBUG datasets
Steps
Download the listed datasets from the facial-point-annotations resource and extract them under /data/images/ in separate 300W, lfpw, helen, afw and ibug directories.
Run TestSetPreparation.py.
Run DANtesting.py to calculate the documented evaluation metrics.
Expected outcome
The evaluation script calculates average error across test subsets, AUC@0.08 and failure rate for the 300W public and private test sets.
Guide limitations
- Dataset acquisition and script execution are described by the README but are not verified here.
- The README also documents a separate Menpo path through MenpoEval.py; this recipe covers the listed multi-dataset comparison workflow.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- MarekKowalski/DeepAlignmentNetwork · Official homepage title · retained in its source language
Related options
zjjMaiMai TensorFlow implementation — A separate TensorFlow implementation of Deep Alignment Network linked by the repository README.
mariolew TensorFlow implementation — A separate TensorFlow implementation of Deep Alignment Network linked by the repository README.