PFLD Tensorflow 2

Implementation of PFLD A Practical Facial Landmark Detector by Tensorflow 2.

Usage

Requirements: tensorflow >= 2.0.0, numpy, opencv, pytorch (optional)
Datasets

WFLW Dataset Download

Wider Facial Landmarks in-the-wild (WFLW) is a new proposed face dataset. It contains 10000 faces (7500 for training and 2500 for testing) with 98 fully manual annotated landmarks.

WFLW Training and Testing images Google Drive Baidu Drive

WFLW Face Annotations

Unzip above two packages and put them on ./data/WFLW/

move Mirror98.txt to WFLW/WFLW_annotations

$ cd data

$ python3 SetPreparation.py

Generate tfrecord:
```
python tools/generate_tfrecord.py
```
Train

You can change configurations in train.py and config.py. For training, just execute one line code.
```
python train.py
```
Test

Just read test.py and load weight you want.
Camera test

You should check if pytorch is installed, and load weight you want.

For loss function, attributes_w_n may all be zero, which makes loss equal to zero. So it may need to rethink about the weight.
We provide a model called PFLD_wing_loss_fn which uses wing_loss and removes auxiliarynet.
A model called PFLD_Ultralight is available, which uses GhostBottleneck.