This Space use the Apache 2.0 Licensed Mediapipe FaceLandmarker
This service offers two object detection methods:object detection using MediaPipe with a selected center point, and object detection using a trained model.
"Current MediaPipe face-landmark detection struggle with faces rotated more than 45 degrees (due to limitations in training data).
Accurate detection in MediaPipe requires correct positioning, but I don't know the exact position needed. This is a known issue
center index see mediapipe face index image
We trained the model using data generated by rotating frontal views(Flux.1 schnell). Because of this rotation, the center points might not align perfectly, potentially causing a shift in the object detection
I'll share model and dataset soon.
