Sorry, me again. I'm trying to figure out the time-lapse feature and I may have completely misunderstood. What I'm trying to do is rather than me manually clicking save on the preview every couple of 100 iterations, is to have this automatically done so that at the end of the training I can produce a video showing how the network learns.
Is this the right feature to do that? If so, will it decrease training speed as from the tooltips it looks like it takes the input photos for each model again and converts them again?
Sorry if this is the wrong place for this question!
You can either point the TimelapseA/B at your input folders, in which case it will take the first (14?) images from that folder and create a timelapse, or you can put selected photos from your trainset into another folder purely for the timelapse.
At every save iteration it will pass these faces through the model and save out an image for those faces into your specified output folder.
It will have minimal impact on your training time. Inference is quick, I doubt you'd even notice the difference.
You can either point the TimelapseA/B at your input folders, in which case it will take the first (14?) images from that folder and create a timelapse, or you can put selected photos from your trainset into another folder purely for the timelapse.
At every save iteration it will pass these faces through the model and save out an image for those faces into your specified output folder.
It will have minimal impact on your training time. Inference is quick, I doubt you'd even notice the difference.
