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.

Faceset Batch-Sharpening with FastStone Image Viewer

Nickrus

DF Admirer
What I'm missing in the deepfakes are sharp faces. I think about applying sharpening tools, it might improve the overall quality.

What can be sharpened:
  1. source facesets (aligned), 
  2. destination facesets (aligned),
  3. processed frames.
First two could help neural network to detect face elements, the third could help a better fap :D

I suggest trying the sharpening and tell if the results are better.

How to try:
  1. Download FastStone Image Viewer here (free full-functional version): https://www.faststone.org/FSIVDownload.htm (Windows only)
  2. Go to the source / destination "aligned" folder, 
  3. Create a folder "aligned_sharpened" for the output images. It can be a subfolder or a folder near the "aligned", doesn't matter. You will choose this path in the settings later.
  4. Go to the "aligned" folder. Open any image with the FastStone Viewer. When in view mode, press enter to switch to Image Explorer view.
  5. Select all images in the folder. Or only the ones that look blurry. Look at how many are there end remind the number.
  6. Select Tools -> Batch Convert Selected Images (F3). Or just hit F3.
  7. Make sure you chose "aligned_sharpened" from step 2 as a destination folder. Then your souce files won't be overwritten.
  8. Check that all the files you wanted to have processed are in the input list. Remember how many should have been there in the source. Add files with a button if necessary.
  9. Check box Use Advanced Options. (see Attachment 1).

    6iuTEcLh.png

  10. Click Advanced Options button. On the Adjustments Tab check box and enter settings as in the Attachment 2. Press OK.

    4cXq0EHh.png

    You can also press Design and Preview to see the difference.

    Example:
    4ZnWTSfh.png

    You can play with the settings to adjust for the better results. I choose 78 / 0.4.
  11. Ok, ready to go! When back in the Conversion window, press Convert and wait. Check the result.
USM (Unsharp Mask) is an intelligent sharpening method, I believe it really improves images. Especially when a human has chosen optimal settings, it is like countouring by hand. I wish that DFL would apply such a setting to every replaced face in the output, it would improve the quality a lot.
 

SPT

Moderator
Staff member
Moderator
Verified Video Creator
Another solution for this is using a video editing software (Premiere Pro, DaVinci, Vegas...) sharpen and color grade src and dst video, and then do the usual stuff with DFL. It's probably quicker or at least similar to your method.
 

dpfks

DF Enthusiast
Staff member
Administrator
Verified Video Creator
Moved to guides. Thanks for sharing.
 

Nickrus

DF Admirer
That is correct, Adobe AE is also good. I wrote this for the guys who don't make DF by themselves buy only extract and upload facesets.


The tool I suggest is compact, simple and free to use, while Adobe AE is heavy, complicated, and expensive.

Also, when you process faceset images, you only deal with the face pixels and in AE it's about the whole frame that can be X times bigger.

The choice is yours.
 

RockNDollars

DF Vagrant
Wouldn't tools like Topaz Sharpen AI be even more efficient? After all if people create deepfake vids they're already supposed to own a fast gpu.  :shy:
 
Top