InsightFace PyTorch Inference

Taken from InsightFace_PyTorch. More explanation and references can be referred to the original repo.

The purpose of this repo is to fix and modify the inferencing section of the original repo.

Tested on Ubuntu 18.04

How to use

Environment setup

1. Clone the repo
2. Download pretrained_model from original repo. Tested with IR-SE50 model

• Download model

3. Place downloaded model in the work_space/save/ directory and rename to model_cpu_final.pth for cpu machine.
4. Install all dependencies from requirements.txt

• Setup virtual environment virtualenv env -p python3
• pip install -r requirements.txt

Prepare data

All face data (images, embeddings, etc) are stored in the facebank/ directory. You will need to make the directory in data/

Then, prepare the face you want to detect. There are 2 preparation methods.

1. Use pre-existing face images by placing them into the directory

data/facebank/
            name1/
                photo1.jpg
                photo2.jpg
                ...
            name2/
                photo1.jpg
                photo2.jpg

2. Use webcam to capture new images

python take_pic.py -n name
press 't' to capture and 'q' to exit

This will create a directory called name in data/facebank. All captured faces are stored inside it.

Test model

python face_verify.py

To update facebank, use -u flag

python face_verify.py -u

Press 'q' to exit.