How do I preload retinaFace and other models?

[martinenkoEduard]

I want to build a simple REST API for face recognition. Is there a way to preload retinaFace and other models beforehand (before first request).

[martinenkoEduard]

What is the best way to use deepface with flask?

[serengil]

yes you can. load the pre-trained model in the following file and pass to analyze or verify funciton.

https://github.com/serengil/deepface/blob/master/api/api.py#L104

[serengil]

i am using flask with gunicorn

[martinenkoEduard]

I mean - that in this case the first call of analyze on this particular gunicorn worker will take a long time, because it will be loading model into memory.

How do I load model when the flask application is starting. So even the first request is fast.

[serengil]

i mean that you can move loading pre-trained model logic in start up

if you change the line 104 as:

models = {}
models['emotion'] = build_model('Emotion')
models['age'] = build_model('Age')
models['gender'] = build_model('Gender')
models['race'] = build_model('Race')
resp_obj = DeepFace.analyze(instances, actions = ["age", "gender", "emotion", "race"], models = models)

this will spend time on app start up and your first request will be fast.

if you do not pass models to analyze function, then the library spend time to build models in the first request.

[martinenkoEduard]

What models should I use for DeepFace.represent and detect_faces?

[serengil]

facenet and retinaface

[martinenkoEduard]

I mean how do I define it?

models = {} models['emotion'] = build_model('Emotion') models['age'] = build_model('Age') models['gender'] = build_model('Gender') models['race'] = build_model('Race')

models['represent'] = build_model('facenet512 ')?

[serengil]

this code snippet is for facial analysis. facenet is face recognition model. they are different.

you can find how to use it in the code docs: https://github.com/serengil/deepface/blob/master/deepface/DeepFace.py#L271

models = {}

models['emotion'] = build_model('Emotion')
models['age'] = build_model('Age')
models['gender'] = build_model('Gender')
models['race'] = build_model('Race')

resp_obj = DeepFace.analyze("img1.jpg", 
                        actions = ["age", "gender", "emotion", "race"], 
                        models = models, 
                        detector_backend = "retinaface"
)