Installers
-
Installer zip file shared here : https://www.patreon.com/posts/121335747
-
This APP is based on DeepFace
https://github.com/serengil/deepface
It is fully multi-threaded and higher batch size increase your processing speed
Make sure to have accurately installed CUDA and cuDNN for TensorFlow library to use your GPU or else it will be ultra slow
It uses minimal amount of VRAM so should work on every GPU pretty much
What this APP does is that, it takes source images folder, target images folder and optionally extracted faces folder
Then it calculates every target image similarity / distance to the source images individually (if multiple source images used it averages them) and at the end sort all target images according to their similarity / distance
When you use more than 1 image in source images folder, it will sort according to the average of the source images but be careful it will increase processing time linearly since each image compared with each target image
If few images gives error at first run, you can run again and it usually fixes error occurred images - i am still not sure why this happens - probably related to DeepFace library
This app can be extremely useful to sort AI generated images according to the original training images to get the best ones comparatively
The images in target folder will be renamed as the followings
Image_Similarity_Rank - Image_Similarity_Score - Older File Name
e.g. image_rock.png will become 33 - 0.55 - image_rock.png
33 is similarity rank, 0.55 is its distance (lesser is more similar)
Following DeepFace models supported but I didn't test them except Facenet512 which is best
Author Of article : Furkan Gözükara Read full article
