deep-head-pose
deep-head-pose copied to clipboard
natanielruiz
Run command field information?
python code / test_on_video_dockerface.py --snapshot ./hopenet_robust_alpha1.pkl --video ./video/1.mp4 --bboxes FACE_BOUNDING_BOX_ANNOTATIONS --output_string new_output --n_frames 2000 --fps 20
-
please, Parameters:FACE_BOUNDING_BOX_ANNOTATIONS How to Write？
-
error info: from torch.utils.serialization import load_lua ImportError: No module named serialization
torch is version?
@natanielruiz please help me with "Parameters:FACE_BOUNDING_BOX_ANNOTATIONS How to Write？"
It's an annotation file, If you have a custom face detector other than dlib you can generate the bounding boxes and save them into a file. If you're not sure, just use the dlib model.
python code/test_on_video_dlib.py --snapshot PATH_OF_SNAPSHOT --face_model PATH_OF_DLIB_MODEL --video PATH_OF_VIDEO --output_string STRING_TO_APPEND_TO_OUTPUT --n_frames N_OF_FRAMES_TO_PROCESS --fps FPS_OF_SOURCE_VIDEO
Download the dlib face model extract it and point the .dat file as PATH_OF_DLIB_MODEL
@sumsuddin thanks for the help, but can this project do head pose estimation in real time on my face through a camera, instead of uploading a video file. The project I want to use it on outputs predictions on a person sitting in front of camera.
Thank You
You can add any webcam using opencv and get the prediction with just a little bit of tweaking.
But in my experience, it's a bit slow to be fully realtime, at least on my Nvidia-1050ti GPU it was about 5 to 6 FPS as far as I remember.
@sumsuddin thanks for such quick response, I am AMD card user so is/are there any other alternative to this project? All I want is the head pose estimation to be as accurate as possible and in real time as my calculations depend on it.
Thank You
Can you share your --bboxes FACE_BOUNDING_BOX_ANNOTATIONS files?
Thank You
