Training with Original trainer: after 35+ k iterations only the 1st A and 1st B images are showing

[cavay]

Hi, I'm a newbie in this so I'm confused as to why even after 35+ k only the 1st images of both sides are showing? I must admit that the next 2 on both sides are slowly showing some pixelated color but the rest is completely blank (grey).

I tried initially the RealFace plugin and there all the images in the preview started blurry.

What should I expect here towards the end (which is far far away, I know)? Will each image (top to bottom) slowly start to appear? If so, does that mean that the current ones that I can see at the top have actually most progressed and the ones that have blanks made no progress?

Note: on every preview refresh the images at the top are different ones

[torzdf]

Hi, I'm a newbie in this so I'm confused as to why even after 35+ k only the 1st images of both sides are showing? I must admit that the next 2 on both sides are slowly showing some pixelated color but the rest is completely blank (grey).

This sounds like one half of the model has collapsed... The two sides should resolve at approximately the same speed.

I tried initially the RealFace plugin and there all the images in the preview started blurry.

I'm not a big fan of the unbalanced models, I would suggest choosing a different one. Dfaker is a reasonable choice if starting out + learning the ropes.

What should I expect here towards the end (which is far far away, I know)? Will each image (top to bottom) slowly start to appear? If so, does that mean that the current ones that I can see at the top have actually most progressed and the ones that have blanks made no progress?

See above

[cavay]

Hi, I'm a newbie in this so I'm confused as to why even after 35+ k only the 1st images of both sides are showing? I must admit that the next 2 on both sides are slowly showing some pixelated color but the rest is completely blank (grey).

This sounds like one half of the model has collapsed... The two sides should resolve at approximately the same speed.

Hmm, not sure I explained well enough so let me try again.

In the preview, there are multiple columns so it looks like this:
(+) : there is an image
(pb) : pixelated background where an image should be
(sb) : solid (gray) background where an image should be

It seems that "pb" are appearing with the increase of iterations. Should I expect at some considerably higher numbers of iterations remaining images to start appearing (I assume first PBs, then SBs)?

Another thing is that on each preview refresh (automatic or manual), these images that do appear (+) - they get changed but apparently within some small set of images, I don't see there a great variety of images.

Original (A)	Original > Original	Original > Swap	Original (A)	Original > Original	Original > Swap	Swap (B)	Swap > Swap	Swap > Original	Swap (B)	Swap > Swap	Swap > Original
+	+	+	+	pb	sb	+	+	+	+	sb	pb
+	pb	sb	+	pb	sb	+	sb	pb	+	sb	pb
+	pb	sb	+	pb	sb	+	sb	pb	+	sb	pb
+	pb	sb	+	pb	sb	+	sb	pb	+	sb	pb

[torzdf]

A screengrab would make this easier... But it should start resembling faces fairly quickly (within a save iteration or 2). If it isn't, then the model has collapsed and you'll most likely need to start again with a lower learning rate.