Skip to content

Latest commit

 

History

History
67 lines (56 loc) · 2.64 KB

README.md

File metadata and controls

67 lines (56 loc) · 2.64 KB
Raw

1. Installation

First, activate the virtual environment:

source ../venv/bin/activate

Then, clone the FaceX-Zoo repository:

git clone https://github.com/JDAI-CV/FaceX-Zoo.git

Now, please do the following commands:

mv extract_frames.py FaceX-Zoo/face_sdk/
mv detect_and_align.py FaceX-Zoo/face_sdk/
mv extract_feature.py FaceX-Zoo/test_protocol/
mv eval.py FaceX-Zoo/test_protocol/
mv backbone_conf.yaml FaceX-Zoo/test_protocol/

FaceX-Zoo checkpoints are available in the following link. Here is the model we used:

https://drive.google.com/drive/folders/1h_meJetsaVUm-37Wqo-o3ed9lyWcS8-B

Download and place Epoch_17.pt in ./FaceX-Zoo/test_protocol/.

2. Usage

2.1 Extract frames from .mp4 files

cd FaceX-Zoo/face_sdk/
python extract_frames.py [--input_root] [--out_root] [--num_frames]

Where --input_root is the directory containing the .mp4 files, --out_root is the output directory, and --num_frames is the number of frames to extract.

2.2 Detect and align

For extracting ROIs and align them, run the following command:

cd FaceX-Zoo/face_sdk/
python detect_and_align.py [--input_root] [--out_root]

Where --input_root is the directory containing the extracted frames, and --out_root is the output directory.

2.3 Feature extraction

For extracting face representations, run the following command:

cd FaceX-Zoo/test_protocol/
python extract_feature.py [--input_root]

Where --input_root is the directory containing the extracted frames after detection and alignment.

2.4 Evaluation

The evaluation stage assumes that the user has already obtained:

  1. .npy file containing the extracted features of a claimed identity reference set (e.g. Barack Obama images).
  2. .npy file containing the extracted features of a test set containing the observed identity and a claimed identity (e.g. images of attacks on Obama and authentic Obama images).
  3. .npy file containing the labels of each sample in the test set.

Finally, you can evaluate by running the following command:

cd FaceX-Zoo/test_protocol/
python eval.py [--real_root] [--test_root] [--labels_root]

Where --real_root is the path to reference set .npy file, --test_root is the path to test set .npy file, and --labels_root is the path to test corresponding labels .npy file.

FaceX-Zoo

The implementation we offer is based upon FaceX-Zoo and its implementation. If you have any implementation-related questions, please refer to the original FaceX-Zoo repository.