I am looking for all you know, just for comparison. I am not looking for methods, but for ready-to-use software which includes the trained models.
What I know so far:
- IntraFace: http://www.humansensing.cs.cmu.edu/intraface/
- Chehra: https://sites.google.com/site/chehrahome/home
- Robust Discriminative Response Map Fitting: http://ibug.doc.ic.ac.uk/resources/drmf-matlab-code-cvpr-2013/
- flandmark: http://cmp.felk.cvut.cz/~uricamic/flandmark/
- Face Alignment at 3000 (reimplementation): https://github.com/jwyang/face-alignment
There is also the tracker by Jason Saragih: https://github.com/kylemcdonald/FaceTracker
and a relatively new publication on ECCV'14 titled: Coarse-to-Fine Auto-Encoder Networks for Real-Time Face Alignment (code and paper can be downloaded here: http://vipl.ict.ac.cn/resources/codes (CFAN))
You might also find something on the result page of the first "300 faces in the wild" challenge, which dealt with accurate location of facial landmarks: http://ibug.doc.ic.ac.uk/resources/300-W/
Hope that helps.
There is also STASM, which is an active shape model based approach which locate several face landmarks on an input face image:
http://www.milbo.users.sonic.net/stasm/
"Project-Out Cascaded Regression" was presented at CVPR 2015. The paper and matlab software (with trained model but no face detection) are available at the website of the author.
http://www.cs.nott.ac.uk/~yzt/
An implementation of the CVPR 2014 paper "One Millisecond Face Alignment with an Ensemble of Regression Trees" is available in dlib, including a pretrained model and code for training own models.
http://www.csc.kth.se/~vahidk/face_ert.html
http://blog.dlib.net/2014/08/real-time-face-pose-estimation.html
We use different methods that were published as open source software. You can have a look at the github pages and the mehods' respective research papers below. Implementations are in C++.
Supervised descent method implementation and sample applications for facial landmark detection:
http://patrikhuber.github.io/superviseddescent/
Adaptive tracking for arbitrary objects: http://adaptivetracking.github.io/
Lightweight 3D morphable face model fitting library: http://patrikhuber.github.io/eos/
Supervised descent and 3D morphable model fitting will be presented together at ICIP 2015 + CppCon 2015, for more information refer to http://www.patrikhuber.ch/
Another Option: http://cmp.felk.cvut.cz/~uricamic/clandmark/
Paper: http://www.sciencedirect.com/science/article/pii/S0262885616300105
Joint Cascade Face Detection and Alignment
https://github.com/luoyetx/JDA
http://link.springer.com/chapter/10.1007/978-3-319-10599-4_8
I personally use OpenFace
https://www.cl.cam.ac.uk/~tb346/res/openface.html
It is free, doesn't even need installation, nice GUI, and has a high success rate!
Cascaded Continuous Regression for Real-time Incremental Face Tracking
https://continuousregression.wordpress.com/
http://link.springer.com/chapter/10.1007/978-3-319-46484-8_39
Adrian Bulat and Georgios Tzimiropoulos: "How far are we from solving the 2D & 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)", ICCV 2017
https://www.adrianbulat.com/face-alignment
For more recent results you can take a look to our facial landmarks detectors:
a) 3DDE with a CNN + Gradient Boosted trees. See code at https://github.com/bobetocalo/bobetocalo_eccv18 (ECCV 2018 and CVIU 2019).
b) CHR2C with a CNN + CNN + heatmap to coordinates layer. It is trained in such a way that the second CNN deals with occlusions. See code at https://github.com/bobetocalo/bobetocalo_prl19 (Pattern Recognition Letters, 2020)
Both detectors have different strengths: the CNN+Gradient Boosted trees is able to adjust the face shape and so is very good at 300W with easier faces (see 3DDE at https://paperswithcode.com/sota/face-alignment-on-300w) and the CNN+CNN is very good with any kind of nuisance (see CHR2C at https://paperswithcode.com/sota/face-alignment-on-wflw).
'openFACE' is a good choice. It is automated and gives out 68 landmarks which you can reduce according to areas of interest based on your objective.
The best landmarks detector on WFLW (without pretraining on large datasets) in 2022 is our SPiGA model (BMVC 2022):
https://github.com/andresprados/SPIGA
It uses a top performing cascade of cnns together with a cascade regressor of graph attention networks (GATs).
There are several publicly available facial landmark detection/tracking software libraries and frameworks that are commonly used for research purposes. Here are some popular options:
- Badges
- Science topic
- Similar topics
- Analytical Chemistry
- Extraction