RetinaFace

0.0.2 verified Sat May 09 auth: no python

A Python implementation of RetinaFace, a single-stage dense face localisation model for face detection and landmark localisation in the wild. Current version 0.0.2, requires Python >=3.6, uses PyTorch. The library wraps a pre-trained ResNet50-based RetinaFace model. Release cadence: low (last update likely 2021).

pip install retinaface-py

Common errors

error ModuleNotFoundError: No module named 'retinaface' ↓

cause Incorrect module name used in import. The correct module is 'RetinaFace' with capital letters.

fix

Use: from RetinaFace import RetinaFace

error TypeError: 'RetinaFace' object is not callable ↓

cause Trying to call the class directly instead of using the static method detect_faces.

fix

Use: detector = RetinaFace(); detections = detector.detect_faces(img) or simply RetinaFace.detect_faces(img)

error ValueError: The truth value of an array with more than one element is ambiguous ↓

cause Passing an image with non-RGB channels (e.g., RGBA) or incorrect dtype.

fix

Ensure image is a 3-channel RGB numpy array with dtype uint8.

Warnings

gotcha The module name is case-sensitive: 'RetinaFace' not 'retinaface'. Import as 'from RetinaFace import RetinaFace'. ↓

fix Use the correct import statement: from RetinaFace import RetinaFace

gotcha Input image must be a numpy array in RGB format, not BGR. OpenCV loads images as BGR by default. ↓

fix Convert BGR to RGB before passing: img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

deprecated The library is no longer actively maintained; last version 0.0.2 released around 2021. May not work with newer PyTorch versions. ↓

fix Consider using alternative face detection libraries like DeepFace, MTCNN, or dlib.

Imports

RetinaFace
wrong
```
from retinaface import RetinaFace
```
correct
```
from RetinaFace import RetinaFace
```
Common mistake: using lowercase 'retinaface'. The correct module name is 'RetinaFace' with capital R and F.

Quickstart

Detect faces in an image using RetinaFace. The input image should be a numpy array (RGB). Returns a dictionary of detections.


from RetinaFace import RetinaFace
import cv2

# Load image (assuming opencv-python is installed)
img = cv2.imread('face.jpg')
img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# Detect faces
detections = RetinaFace.detect_faces(img_rgb)

# Print results
for face_id, face_data in detections.items():
    print(f"{face_id}: score={face_data['score']}, box={face_data['facial_area']}")