Hello, does this model heavily rely on the pre existing face detection model? If the face detection model is not very robust, will there be any issues with the model? I found during the experiment that the effectiveness of face detection greatly affects the effectiveness of the facial keypoint localization model. Looking forward to your reply #56
Yes, I think the detection model will affect the landmark model. In short, you need to make sure the landmark model is trained with images that cropped the same way as the detection model you use for inference. And different detection model may have different bounding box styles, e.g., some give you more boundaries and some are quite tight, etc.
Yes, I think the detection model will affect the landmark model. In short, you need to make sure the landmark model is trained with images that cropped the same way as the detection model you use for inference. And different detection model may have different bounding box styles, e.g., some give you more boundaries and some are quite tight, etc.