Face Detector in MediaPipe model maker

TensorFlow General Discussion
, ,
1

I see that face detection is missing in MediaPipe Model Maker and I was wondering whether it will be supported any time soon. Is there a roadmap for MediaPipe Model Maker that can be shared?

3

Hi @Jason_Xie, AFAIK the face detection task can be achieved using MediaPipe model maker object detection task. Thank You.

5

Thank you! I will give that a try.

7

Hi, do you know what the format of the annotations needs to be in order to detect the face landmarks in the face detection task? I used the COCO format for the training and validation data.

9

Would anyone be able to help with me training a mediapipe object detector to detect both the bounding box and keypoints?

I used the COCO dataset format and I have set up the labels.json to be as follows:

{
	"image_id": 1,
	"category_id": 1,
	"bbox": [x, y, w, h],
	"num_keypoints": 5,
	"keypoints": [x1, y1, x2, y2, x3, y3, x4, y4, x5, y5]
}

When I run the resulting object detector on images I get the bounding box but no keypoints. The detection result does contain the keypoints attribute but it is empty.
Does anyone know what I am missing?