Open Tymon555 opened 4 years ago
Well, they are currently fully ignored... I was thinking about what we could do for video is do some interpolation somehow if there is a face missing in a frame. Are you getting a lot of NaN's?
I haven't measured what is the proportion of NaNs. Might be worth trying
I've been looking through it and it seemed like there were almost no NaN's. They would've been eliminated during FDHH or if not there would be an error raised
https://github.com/OvidiuGrec/Darwin-Project/blob/8956f1b41a8212949cf03da481e82bd16790fe9e/video_features.py#L40
Does anyone know how NaN values are treated when extracting the features? Do you think it may influence the results?