MrDeepFakes Forums

Some content may not be available to Guests. Consider registering an account to enjoy unrestricted access to guides, support and tools

  • We are looking for community members who are intested in helping out. See our HELP WANTED post.

What to do when the manual face extraction fails to capture the face

TS6lKySh.png


This is one of the frames from 4k quality porn for dst files. 
We can probably recognize there is a face there, and kinda distinguish where her eyes, nose, and mouth is.
However, the manual extraction tool doesn't seem to grasp any parts of the face.. I have tried for last 20 minutes trying to align these
kind of photos, but to no avail. Has anyone had same problems? Any way to solve this?
I know the easy solution is to just discard the video and get a new part with static face, but that doesn't seem to fun
 

Nickrus

DF Admirer
You know everything already: software needs steady and crispy faces, that's it. Try removing the most dynamic parts. Or wait for a hist from someone experienced (I'm not).
 
Nickrus said:
You know everything already: software needs steady and crispy faces, that's it. Try removing the most dynamic parts. Or wait for a hist from someone experienced (I'm not).


Thx for the reply! I just decided to abandon the video and get a new one. such a shame though


Nickrus said:
You know everything already: software needs steady and crispy faces, that's it. Try removing the most dynamic parts. Or wait for a hist from someone experienced (I'm not).

Well.. this is from Jav series, named as VDD-XXX. As all video here, I doubt he knows, and I don't really know him either with that mosaic on
 

LCC

DF Pleb
Boom, sharpened it up for you in PS.

t9HtMKph.png


If that doesn't work, post the frame directly before and after it and I'll try something else for you.
 
LCC said:
Boom, sharpened it up for you in PS.

t9HtMKph.png


If that doesn't work, post the frame directly before and after it and I'll try something else for you.
OMG
I greatly appreciate your help. It finally worked XD. I guess PS may be the way to go in these kind of cases
 

dpfks

DF Enthusiast
Staff member
Administrator
Verified Video Creator
This is interesting. I am testing sharpening images into my workflow.

I will post a guide if it works out.
 

titan_rw

DF Pleb
Hmm. I wonder if it's advantageous to intentionally over-sharpen the DST frames, then detect and extract faces. But revert to the original non sharp images for conversion. So the NN is working off of 'sharp' images that might not please our eye, but might work better for machine learning, whereas the actual converted images come from images that please us.

It is after all, all about the network learning face 'features', maybe an over-sharpened image is better for that.
 
titan_rw said:
Hmm.  I wonder if it's advantageous to intentionally over-sharpen the DST frames, then detect and extract faces.  But revert to the original non sharp images for conversion.  So the NN is working off of 'sharp' images that might not please our eye, but might work better for machine learning, whereas the actual converted images come from images that please us.

It is after all, all about the network learning face 'features', maybe an over-sharpened image is better for that.

Is that even possible?  Whenever I tried stuff like that I could not use the aligned extracted faces from the sharpened/tweaked images with the original frames.  The conversion just failed and I was left with the original frames in the merged folder with no src face.  Maybe my save settings messed something up.  What is the best software for this?





I used top secret techniques to enhance the face of Mr. Blurry Face... the results were... surprising... 

l4kiDcLh.jpg
 

LCC

DF Pleb
OMG

I greatly appreciate your help. It finally worked XD. I guess PS may be the way to go in these kind of cases

Glad it worked out for you mate.
If anyone else has this problem, the method I used was Photoshops Shake Reduction sharpen (Youtube has some good tutorials). Then using an inverted layer mask over the result, paint in the features for tracking (e.g. face edges, eyes, mouth, eyebrows).

This is interesting. I am testing sharpening images into my workflow.

I will post a guide if it works out.

Would love to see your workflow on this if it's successful. Adding a bit of sharpening while colour correcting source videos, but it's nowhere near as good as PS.
Lightroom might be a better option to sharpen large batches of PNGs before extracting the faces. But I have waaaaaay less experience with it.
 
Top