(unofficial) DFL 2.0 GUI (user interface)

[angeloshredder]

Oh , that's great job , especially if done by a non-developer , I downloaded the last version , it's compatible with DFL 2.5.
Really very helpful especially in merging , interactive merger was annoying for me , but now it's easy 

Thanks man  :heart:
And keep the good working 
---------
if you can make a manual user it will be perfect . especially for merger part ( it was difficult for me to understand how to save merger-settings to apply them later)

 

[Roy_B]

Oh , that's great job , especially if done by a non-developer , I downloaded the last version , it's compatible with DFL 2.5.
Really very helpful especially in merging , interactive merger was annoying for me , but now it's easy 

Thanks man  :heart:
And keep the good working 
---------
if you can make a manual user it will be perfect . especially for merger part ( it was difficult for me to understand how to save merger-settings to apply them later)

Ciao!

I'm newbie at DFL, so i recommend to read the official guide here: in that page there's also a useful  that explain the merging workflow quite well

 

[hola2]

can you enable gan power to go a lower value like 0.01? lowest value it accepts is 0.1.

 

[Roy_B]

can you enable gan power to go a lower value like 0.01? lowest value it accepts is 0.1.

If you use the "manual input" button (or use DFL normally) you can manually enter any value that you want (like 0.01): it will be displayed on the training summary, but i really don't know if DFL actually accepts that value "as it is" or approximates it to 0.0 or 0.1 when processing data

You'd better ask iperov for that

 

[hola2]

can you enable gan power to go a lower value like 0.01? lowest value it accepts is 0.1.

If you use the "manual input" button (or use DFL normally) you can manually enter any value that you want (like 0.01): it will be displayed on the training summary, but i really don't know if DFL actually accepts that value "as it is" or approximates it to 0.0 or 0.1 when processing data

 You'd better ask iperov for that

Ohh it works under 0.1.I use it.Sometimes even 0.1 has a big effect.I tried manual input field on your interface but i wasn't able to put 0.01 there but infact if you use dfl without interface with bat files you can put values 0.04 - 0.05 and of course the gan power applied is reduced.That is why I asked for it. You can check it if you want.
Also thank you for the interface and your work.

 

[Roy_B]

I tried manual input field on your interface but i wasn't able to put 0.01 there

I've just made a quick test to verify manual input mode and it worked for me

P.S. anyway i've updated the gui to support double decimals for GAN

 

[TMBDF]

Forgot about this, it's nice that you're keeping it updated, btw can you specify folders where models and datasets are or do you still need to place them in the workspace folder with this GUI?

 

[Roy_B]

Forgot about this, it's nice that you're keeping it updated, btw can you specify folders where models and datasets are or do you still need to place them in the workspace folder with this GUI?

Unfortunately no, since the gui simply starts the DFL batch files, it does not modify/use their content to adapt them for custom paths.

Personally for convenience i create multiple "workspace" subfolders inside DFL (like workspace_1,2 etc.) and rename them when needed

 

[TMBDF]

Yeah but it's not good for your drives, repeated copying/pasting of big datasets and models uses them up, especially SSDs that have limited save cycles but also HDDs.

 

[Roy_B]

Yeah but it's not good for your drives, repeated copying/pasting of big datasets and models uses them up, especially SSDs that have limited save cycles but also HDDs.

You mean that's better to keep workspace folders or data_src/dst folders separated? Even better if on different SSD/HDDs?

 

[angeloshredder]

Most SSD begin their failure after writing 700-800 TB, some of them can go for more than 2 FB ( 2000 tb) , so if you have the worst SSD in the world ( a 700 TB writing life cycle)  and you copy 100GB /day then your SSD Will not die for 7000 days = 19 years 
you can check your Hard-Drive health by using a free tool

 

[TMBDF]

Most SSD begin their failure after writing 700-800 TB, some of them can go for more than 2 FB ( 2000 tb) , so if you have the worst SSD in the world ( a 700 TB writing life cycle)  and you copy 100GB /day then your SSD Will not die for 7000 days = 19 years 
you can check your Hard-Drive health by using a free tool

The values you mentioned are not valid and they differ from manufacturer to manufacturer of NAND chips used (and often are lower for cheaper/budget SSDs), usually a company that makes your SSD will also provide a spreadsheet with MTBF (Mean Time Before Failure, in hours) as well as TBW (Total Bytes Written, in TB) which not only differ depending on the NAND chip used (SLC, MLC, TLC) but also on the overall size of the drive (the higher the capacity the higher the TBW value will be). For example my drive is a fairly cheap 240GB TLC NAND based 2,5" SSD and I've already used up about 10-12% of it's write cycles after about 3 years of use (1 year of faking) but it's not the SSD I worry about but the HDD, you repeatably copy your datasets from HDD to SSD to work on fakes, same goes for models and then you copy it all back once it's finished, including any video editing software project files, result and result_masked files and final renders and that's a lot of work for an HDD compared to SSD.


Having the ability to just specify a folder to load from so that we only read the datasets (very good for SSD and a bit better for HDD) would be a nice addition, but then you still save the model every few thousand iterations which would be done directly to SSD or HDD where you keep your model anyway, but at least it would save on additional strain if you copy between archival and workspace SSD and/or HDD.

 

[hola2]

I tried manual input field on your interface but i wasn't able to put 0.01 there

I've just made a quick test to verify manual input mode and it worked for me


P.S. anyway i've updated the gui to support double decimals for GAN

Thank you very muchfor the update,now it accepts the value on change parameters tab.
I have another question if you don't mind.I tried an old 224 res liae model which has 5 preview sections,not like the ones we have now and the preview window is cropped around 3 pictures horizontal, 2,5 pictures vertical.Is there anything I can do to see the full preview window like it is a floating window without UI?

 

[Roy_B]

and 
Thanks for precious info about SSD lifetime. For what it's worth i have an 'old' 128GB Crucial M4 SSD that i use everyday (from years, 35000+ hours) and it's still alive and in good conditions.
I know that it could quit on me at any moment, so i use it only for "disposable" things

Thank you very muchfor the update,now it accepts the value on change parameters tab.
I have another question if you don't mind.I tried an old 224 res liae model which has 5 preview sections,not like the ones we have now and the preview window is cropped around 3 pictures horizontal, 2,5 pictures vertical.Is there anything I can do to see the full preview window like it is a floating window without UI?

Just use the arrow buttons in the top/left corner to shrink and detach the gui from any other app/window

---

Having the ability to just specify a folder to load from so that we only read the datasets (very good for SSD and a bit better for HDD) would be a nice addition, but then you still save the model every few thousand iterations which would be done directly to SSD or HDD where you keep your model anyway, but at least it would save on additional strain if you copy between archival and workspace SSD and/or HDD.

I could try to integrate the necessary batch files into the code and insert a variable that changes according to the user's choice.

To understand each other better, i put an example of the batch file "6) train SAEHD.bat"

 off
call _internal\setenv.bat

"%PYTHON_EXECUTABLE%" "%DFL_ROOT%\main.py" train ^
    --training-data-src-dir "%WORKSPACE%\data_src\aligned" ^
    --training-data-dst-dir "%WORKSPACE%\data_dst\aligned" ^
    --pretraining-data-dir "%INTERNAL%\pretrain_CelebA" ^
    --model-dir "%WORKSPACE%\model" ^
    --model SAEHD

pause

If i can make the program change these two lines on the fly according to user's choice via a "select folder" dialog:

--training-data-src-dir "%WORKSPACE%\data_src\aligned" ^   --->   to something like   --training-data-src-dir "*** SRC FACESET ***" ^
--training-data-src-dir "%WORKSPACE%\data_dst\aligned" ^   --->   to something like   --training-data-dst-dir "*** DST FACESET ***" ^


I don't know how to do it atm but i can try, maybe it could be a solution... any help is welcome

 

[qaz123vs123]

excuse me , I want change the languege of the GUI.Which file should I modify?

 

[Roy_B]

excuse me , I want change the languege of the GUI.Which file should I modify?

English only

 

[SPT]

excuse me , I want change the languege of the GUI.Which file should I modify?

English only

From what I understand, he might want to do a translation of the GUI language. What file should he modify if he were to do that ?

 

[Roy_B]

excuse me , I want change the languege of the GUI.Which file should I modify?

English only

From what I understand, he might want to do a translation of the GUI language. What file should he modify if he were to do that ?

There's no file for language at all, nor the ability to create one (to my knowledge): everything is embedded within the Visual Studio code itself

 

[qaz123vs123]

[quote pid='27139' dateline='1603343643']
qaz123vs123excuse me , I want change the languege of the GUI.Which file should I modify?

English only

From what I understand, he might want to do a translation of the GUI language. What file should he modify if he were to do that ?

There's no file for language at all, nor the ability to create one (to my knowledge): everything is embedded within the Visual Studio code itself
[/quote]

  thank you for your answer.

 

[SPT]

Hi

Been using your GUI for a few weeks, just for merging preview, as it is way better than original interactive preview. Thanks for your work.

One issue I have tho, is that I can't change anything under the SAEHD button, everything is greyed. I would especially like to change the "number of workers" parameters as it's 1 by default on your GUI, but I usually use 8 with standard DFL.

That's not a huge problem, since I just try to find the right parameters with your GUI, then copy the parameters in the bottom left corners and re-use all with normal DFL.

Still I really would like to be able to understand how to change this greyed parameters. I hope you see what I mean, if not I can post a screenshot later, just tell me.


On a totally different subject, if you have time, please review this thread : //deep.whitecatchel.ru/literotica/forums/thread-ideas-for-dfl and check if you can find stuff that you can manage to make happen. I know some of the ideas in there are unfortunately probably only doable for iperov, but there might be some others you can do.
I would especially like to see the first idea (seeing filenames on training and XSeg preview screens) and I would really like to have a keyframe system integrated for merging. But that would also require smooth and fast browsing thru the timeline of the video, which seems like a complicated endeavour.

Anyways, I hope you go on working on it and thanks again for everything you did already.