Install pytorch from pytorch.org and run pip with the requiements.txt
1.) Add a path to videos to PATH2VIDS in DataTool.py (usually data is in the dir /data/). The video must be made by a wide angle camera that shows the face in a very short distance close up. If the videos are rotated, set ROTATE to True. 2.) Run DataTool.py. It loads the video, detect facial landmarks and automatically adjust a bounding box of the lower face. DataTool.py produces a dataset of landmarks based on the bounding box area. 3.) Add the path to the landmark_cropped.csv in train.py around line 36 "completeDataset = ds.FaceLandmarkDataset("Put path here")" 4.) Run train.py