Ideas for DFL

[TheHumbleMrBlue]

Thanks "honor" but I'm using google/colab.

 

[bukabumbum]

Idea for src face extractor: historgram autolevel option for dark images

 

[honor]

Thanks "honor" but I'm using google/colab.

You could Clone... this repo on Google Colab too ( instead of Iperov's ) ( They are identical except the added features )

 

[angeloshredder]

I would like to have the option in training to make the mouth a priority so I can get more detail in teeth (it's especially difficult to get detail in the bottom teeth).

Use this version of DFL - it has mouth Priority and even option for using OLD gan if you want ... 

 

interesting , this guy make some nice repos .. tried his mod and it's working , thanks

 

[TheHumbleMrBlue]

I would like to have the option in training to make the mouth a priority so I can get more detail in teeth (it's especially difficult to get detail in the bottom teeth).

Use this version of DFL - it has mouth Priority and even option for using OLD gan if you want ... 

 

interesting , this guy make some nice repos .. tried his mod and it's working , thanks

Can you teach me how to do that?  :blush:

 

[angeloshredder]

I would like to have the option in training to make the mouth a priority so I can get more detail in teeth (it's especially difficult to get detail in the bottom teeth).

Use this version of DFL - it has mouth Priority and even option for using OLD gan if you want ... 

 

interesting , this guy make some nice repos .. tried his mod and it's working , thanks

Can you teach me how to do that?  :blush:

First you need a copy of the full DFL from iperov repo (the one that you get by using torrents) , then download the zip file from this  , unzip it and copy all the files in the folder "DeepFaceLab-master" to your copy of full DFL and put them in this folder  /_internal/DeepFaceLab
that's all ..
now you have 2 DFLs

 

[TheHumbleMrBlue]

I meant 'how do I clone seranus/DeepFaceLab on colab?' "honor" ? can you help?

 

[honor]

I meant 'how do I clone seranus/DeepFaceLab on colab?' "honor" ? can you help?

I have PM' ed you 

 

[Roy_B]

I think these are rather complicated things to do, and unfortunately i can't put them inside the GUI

- ability to specify folders where models and datasets are, without the need to place them all in the workspace folder (probably someone else have already asked for this)

- possibility to pause/resume "5) data_dst faceset MANUAL RE-EXTRACT DELETED ALIGNED_DEBUG" (useful when there are hundreds of landmarks to re-align, without tha hassle to re-check bad landmarks everytime)

- a (sort of) faceset "analyzer/comparator" to better detect all facial angles/expressions differences between SRC/DST, in order to "fill the gaps" in source facesets and create a proper one that's ready for training

 

[gargamel9]

Getting HD videos for SRC model other than from US actresses is almost imposible. Only a few pics are usually available.
I would be really good it somehow the tool could generate a faceset from a few pics.

 

[C.Rosenkreuz]

Hello, , would it be possible to have filenames like this in the preview ? So that we instantly know the name of the file when we see a wrong image we forgot to delete. Instead of looking inside the whole folder.


Another idea, maybe more complicated to do :




A way to view the images causing these peaks in the graph, so that we can correct or delete them.


What do you think ?

Totally agree, i think they would be very usefull.

 

[SPT]

A way to program DFL a bit like you'd do for trading : Buy if x goes under $100, sell if y goes above $300 etc....

Applied to DFL it would be something like :

Train RW batch 8 until loss numbers are 0.30 or under, then train RW + Yaw for 2h30, then train batch 6 with RW + LRD for 1h etc....


Can't believe a system like this doesn't exist yet. Sometimes you're not home for a long time or sleep more than usual, and we all know that training 15h without changing parameters is either a waste of time or sometimes even harms the model.

So anyone please make this.

 

[SPT]

Hello, , would it be possible to have filenames like this in the preview ? So that we instantly know the name of the file when we see a wrong image we forgot to delete. Instead of looking inside the whole folder.


Another idea, maybe more complicated to do :




A way to view the images causing these peaks in the graph, so that we can correct or delete them.


What do you think ?

7 months anniversary of this request. On a serious note, is DFL made in such a way that someone other than Iperov could code these ideas into it ? (Especially the first one, the 2nd one doesn't matter that much). "Roy_B" "Tuts" (not asking either of you to do it, just asking if you know if anyone besides iperov could do it)

 

[TMBDF]

Yes, you can fork dfl and implement those things yourself, maybe once you code you could submit it to the MVE guy so that he adds it to MVE fork.