You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, I have my face dataset ready and wish not to perform duplicate face detection during inference as it brings huge time cost. However, when I commented out the face detection code and directly ran the inference on my cropped faces (using FFmpeg and dlib) from FF++, it seems that the performance dropped a lot. I wonder if there is a solution to it.
The text was updated successfully, but these errors were encountered:
Hi,
The performance of the classifier strongly relies on the preprocess, e.g., cropping.
We recommend you use the same preprocess as we provide in this repository.
Hi, I have my face dataset ready and wish not to perform duplicate face detection during inference as it brings huge time cost. However, when I commented out the face detection code and directly ran the inference on my cropped faces (using FFmpeg and dlib) from FF++, it seems that the performance dropped a lot. I wonder if there is a solution to it.
The text was updated successfully, but these errors were encountered: