Qhimm.com Forums
Miscellaneous Forums => Tools => Misc. Tools => Topic started by: satsuki on 2019-04-18 21:00:33
-
As this soft was designed to help to start with esrgan, a better one exists here (thanks SlaughterThemAllRichtofen for telling me about it) : https://github.com/n00mkrad/cupscale (https://github.com/n00mkrad/cupscale)
For advanced use never my soft nor cupscale is realy usefull as any advanced esrgan user will do his own processing, but to test and simple use of esrgan cupscale is realy good.
Hello, as i have cleaned the tool i use for esrgan + waifu prefilter (to share it with someone who needed it), I share it here too... may be helpfull for some of you ^^
(http://wowsatsuki.free.fr/115.PNG)
(http://wowsatsuki.free.fr/esrg3.png)
You need to extract the archive then follow the howto before using it (need to install cuda, python...)
You need a cuda9/10 capable gpu (nvidia gt9xx at least)
64bit windows.
The processing can take a lot of time (about 3/5s for a 640*480 picture on my gtx1060)
As ESRgan is realy gpu/ram dependant, you're going to be limited in source picture size (if too big the esrgan won't process depending en your gpu/ram)
I included a 'cpu' processing so you can use it without cuda but it will be realy slow (about 7min for a 640*480 on my core i7 8700k but on small picture it remain resonably fast)
download link : http://wowsatsuki.free.fr/ESRganResizer.1.1.5.7z
-
While I personally cant make much use of this; it is objectively speaking a great tool! Good work.
-
Just wanted to say thank you for the tool. It makes using ESRGAN more practical. I was hoping you implemented some kind of image splitting for when the picture doesn't fit into the available VRAM. Is that something you plan on implementing at some point ?
-
Hmm, looks like Mass Effect gets a new texture improvements.
-
Hello, I downloaded this tool, installed the pre-requisities, installed the python and processed all the scripts succesfully.
I have gtx970, windows 7 x64, 8gb ram
Yet, when I finally click on the ESRganResizer I get:
Unexpected error
Is there any way to find out what is going on? Some logs? Anything?
Thanks in advance!!
-
Can you post a screenshot ?
Have you tried to run it as adminstrator
I haven't tested it in windows 7 but it needs a 64bits os (are you on windows 64 bits or 32 bits ?)
-
Hmm, looks like Mass Effect gets a new texture improvements.
My, my, am I that famous ? :D
I'm looking into ESRGAN and Gigapixel for some more texture improvements, but for now Gigapixel has proved so much easier to use than ESRGAN that I almost only use it. Hence my question about image splitting, which would convince me to give ESRGAN a more serious try. That and the pregenerated models alsa64 shared on Discord, which make my life a lot easier.
-
Version 1.01 on the first message:
-Corrected a bug when you uncheck the waifu prefilter without inchecking the double step.
-Added a security so you can't launch with an empty list.
-Added a 'cpu' mod so you can test waifu + esrgan without a cuda gpu.
-
For some reason i can't get the output file when i render with Cuda.
I have RTX 2080ti with i9-9900k.
-
maybe you can try with cuda 10, don't know if cuda 9 is supported for such a new card
-
That is in fact the reason, rtx needs cuda 10 torch.
install pytorch torchvision cuda100 -c pytorch
-
thanks, command added to the howto in first message
-
I dont have the prog anymore because I was not able to run it.
But the screenshot would not made any benefit, it was a white tab 5x4cm with a text which stated: "Unexpected error" and that was it.
Yes, I tryed running as admin.
And yes, I have 64bit OS (Win7 x64)
Absolutely no idea on what was going on, because there were no logs to hint on the source of a problem :X
Im very sad about it, I work on Daggerfall Remaster, 2nd year, wanted to try to use this prog on some images/sprites :X
Anyway, there is this msg for you from the daggerfall forums form the user MrFlibble, who is active in this field on many forums:
Instead of copy paste, I link you there... 2nd post on that page:
https://forums.dfworkshop.net/viewtopic.php?f=14&p=23696#p23696
Be well! And Im damned :/ Suxx I cant use this tool, why o why
-
Hi there,
I also tried with CUDA 10.0 before. But no luck. But i can share my ESRGan GUI settings and CMD output if it's gonna help you.
CMD Output
(https://i.imgur.com/Ek2Zh1K.jpg)
ESRGAN GUI Settings:
(https://i.imgur.com/ph1KdQd.jpg)
Cheers
-
Seems waifu don't do its job.
Have you tried without waifu prefilter ?
-
Next version on its way :
(http://wowsatsuki.free.fr/w105.png)
With this version you can use it as the current version with provided prefilters, but you'll can also set up your own prefilter commands.
It'll take me some time to do a proper setup but i'll share it asap
-
Yes i've tried literally every possible way but no luck.
-
Looking forward to the next version. There are a few models for de-dithering and JPEG artifact removal that are x1 models. Do you know if you can code this support in? So far just been manually editing the .py file to have upscale=1, or making a copy of that file with equaling one and then swapping file. I think model 1 and model 2 at the moment use the same python file and there is not a way to mix an x1 and x4 file (for obvious reasons). Default should naturally be x4 as that is what the majority of models are setup to be, but there is a growing amount of x1 models I think. There are a few x2 models too that I know of.
-
I'm really looking forward to your next release. Probably going to be using it on all of my mods. :D
-
Looking forward new release and praying it will actually work for me.
Thanks for contacting our community at Daggerfall forums :)
-
@themadpatter
I'll try to implement it in a easy way
@Mcindus
Maybe in a day or two, i'm realy busy ATM
-
Version 1.05 ready:
-Lots of UI change
-Added a way to use 1x,2x,4x,6x and 8x models
-Added a prefiter section so you can use your own prefilter's commands
-Added as save/load command to save and restore your own commands
-Added progress bar for easy check the processing
-Added a log file you can view in case of bugs
....
-
Great work on that update! Things work well for my different models. I seem to be able to run the GUI version of waifu2x-caffe.exe fine, but somehow not able to properly run the commands through waifu2x-caffe-cui.exe. I gotta figure out what might be blocking that for me, but then my setup should work quite well. Thank you!
-
Version 1.06 ready:
-Updated waifu2x-caffe to lastest version
-Added an option to choose cuda or cudnn engine for waifu2x-caffe
-Corrected a bug with noise level reduction with waifu command generator (thanks King of Worms)
-
Swapping to CUDA fixed my issue. Amazing! :D
-
1.07 ready (mostly changed things to easy errors detection):
(http://wowsatsuki.free.fr/err2.png)
-Added error detection in prefilter or esrgan processing
-Added error log to esrgan so you can easily see the file witch bugs
-Modded esrgan script so if a file bugs, the processing continue with other files instead of stopping
-Changed the temporary file name to a source file name + numbers so you can easily bug hunting ^^
-If error detected in processing, the temp files aren't auto removed (as long as you don't launch another processing) so you can check the files.
-
I'm getting an error message that keeps saying the program can't find Python, even though I literally installed Python. Any ideas?
(https://i.imgur.com/y9dVfOa.jpg)
(https://i.imgur.com/vlcNLqG.jpg)
(https://i.imgur.com/uCrZUOv.jpg)
-
have you added python to system path ?
-
1.08 ready:
-Corrected an overflow bug with big batch files
-Added an option to split the source file in 16 smaller pictures before esrgan processing, with this you can process big file you can't process because of gpu ram limits, but it can generate clipping ... for tests ^^
-
have you added python to system path ?
Seems like he did not. I had a same problem.
In the "how to" its stated to install the Python with default settings.
But by default, the "set as system path" option is DISABLED.
That can be confusing for Python newbies like me, also, that setting is easily missed, its not very visible.
It might be worth it, to change the description in HowTo to explicitly state to "enable system path" option.
Just a idea, hope it helps, thanks for a great prog!!!
-
i'm noticing an issue with certain image dimensions causing the tiler to create more than 16 tiles and then when the combiner goes to create the final image it fails. cropping the source image down a pixel or two in either height or width usually fixes it but its an annoying thing to do.
i've also had large batches along with image splitting causing overflows.
-
to use the split in 16 feature you need of course to have a width and height as a multiple of 16 or it won't works for exemple 800*608 will be ok but 800*600 won't
i'll do a more flexible version asap
-
Version 1.09 released :
Changelog
-Corrected a bug if there's special character in folders path
-Added a way better split option for use with esrgan with overlap to avoid bad assembly in final file (and you don't need to have a specific input size to get it works, only a peer number)
-Updated the howto
If you're updatding from older version please install the \runtime\Git-2.21.0-64-bit.exe file
-
I hope to use this tool for easing the use of ESRGAN to sharpen images for the purpose of deepfaking. I know this is a FF related site but I found no other resource to try and resolve my issue. I keep getting a "The system cannot find the file specified" error. Any help would be appreciated. See debug below:
#############################################################
:#0#: Prefilter of 00068.png
#############################################################
C:\ctp\ESRganResizer>cmd.exe /c start /B /wait /low " " "C:\ctp\ESRganResizer\waifuC\waifu2x-caffe-cui.exe" -i "C:\Temp\DeepFaceLabJesus\workspace\data_src\00068.png" -p cudnn -m noise -n 1 -c 128 -t 0 -o "C:\ctp\ESRganResizer\provo\provo.1.00068.png" --model_dir "C:\ctp\ESRganResizer\waifuC\models\photo"
C:\ctp\ESRganResizer>copy "C:\ctp\ESRganResizer\provo\provo.1.00068.png" "C:\ctp\ESRganResizer\esrgan\LR\00068.png"
The system cannot find the file specified.
#############################################################
#############################################################
#############################################################
:# Splitting sources files, can be long, please wait and don't close the commands files #:
#############################################################
#############################################################
#############################################################
C:\ctp\ESRganResizer>C:
C:\ctp\ESRganResizer>cd "C:\ctp\ESRganResizer\"
C:\ctp\ESRganResizer>move "C:\ctp\ESRganResizer\esrgan\LR\*.*" "C:\ctp\ESRganResizer\input\"
The filename, directory name, or volume label syntax is incorrect.
C:\ctp\ESRganResizer>cmd.exe /c start /B /wait /low " " "C:\ctp\ESRganResizer\step1_create_tiles.sh" --output-dir="C:\ctp\ESRganResizer\esrgan\LR"
#############################################################
:##: Starting ESRgan upscale
#############################################################
C:\ctp\ESRganResizer>C:
C:\ctp\ESRganResizer>cd "C:\ctp\ESRganResizer\esrgan\"
C:\ctp\ESRganResizer\esrgan>cmd.exe /c start /B /wait /low " " python test.py models/1xBC1_take2_260850.pth
Model path models/1xBC1_take2_260850.pth.
Testing...
#############################################################
:##: ESRgan upscale end
#############################################################
#############################################################
#############################################################
#############################################################
:# Merging result files, can be long, please wait and don't close the commands files #:
#############################################################
#############################################################
#############################################################
C:\ctp\ESRganResizer\esrgan>C:
C:\ctp\ESRganResizer\esrgan>cd "C:\ctp\ESRganResizer\"
C:\ctp\ESRganResizer>cmd.exe /c start /B /wait /low " " "C:\ctp\ESRganResizer\step3_assemble_tiles.sh" --input-dir="C:\ctp\ESRganResizer\esrgan\results" --input-postfix="_rlt"
C:\ctp\ESRganResizer>del /Q "C:\ctp\ESRganResizer\esrgan\results\*.*"
C:\ctp\ESRganResizer>move "C:\ctp\ESRganResizer\output\*.*" "C:\ctp\ESRganResizer\esrgan\results\"
The filename, directory name, or volume label syntax is incorrect.
#############################################################
:#0#: Postfilter of 00068.png
#############################################################
C:\ctp\ESRganResizer>copy "C:\ctp\ESRganResizer\esrgan\results\00068.png" "C:\Temp\DeepFaceLabJesus\workspace\data_src\00068.final.png"
The system cannot find the file specified.
C:\ctp\ESRganResizer>del /Q "C:\ctp\ESRganResizer\provo\*.*"
C:\ctp\ESRganResizer>del /Q "C:\ctp\ESRganResizer\esrgan\LR\*.*"
C:\ctp\ESRganResizer>del /Q "C:\ctp\ESRganResizer\esrgan\results\*.*"
C:\ctp\ESRganResizer>del /Q "C:\ctp\ESRganResizer\input\*.*"
C:\ctp\ESRganResizer>del /Q "C:\ctp\ESRganResizer\output\*.*"
#############################################################
:##: Finished
#############################################################
-
*resolved*
dropped cudnn64_7.dll in waifuc folder.
-
Satsuki, thanks a lot for this program, it works quite ok even on my problematic PC setup now.
Is there a way to setup the program so it doesnt add the "_Final" suffix to the new files?
Because I have to manually rename them afterwards to the original names...
Thank you :)
-
Not ATM but you can use ant renamer for exemple for batch rename in seconds
-
Thanks for suggestion
I found this utility for batch renaming, its VERY detailed and easy to use, so I will share it, its very handy for tasks like this
https://www.bulkrenameutility.co.uk/Download.php
-
hello, i'm using this tool but when it finishes it gets a badly fitted
Here's how it looks https://imgur.com/00Yyxz1
can you fix this?
sorry for my bad english.
-
hello, i'm using this tool but when it finishes it gets a badly fitted
Here's how it looks https://imgur.com/00Yyxz1
can you fix this?
sorry for my bad english.
Seems like a split issue, what are the settings you use ?
-
Version 1.1.0 released :
Changelog:
-Corrected the split source function.
-Now the processed files keep theirs original name in and are put in an "esrgan" subfolder of the original file folder's.
-You can include the processing steps in the final filename (for exemple a file "test1" will be named "file1.(doublestep).(upresnet10-2-box).(modelSatsuki.v1.pth).(postfilter-0.5)"
-Added a link to the alsa64 esrgan model github to download more esrgan models.
-
Version 1.1.0 released :
Changelog:
-Corrected the split source function.
-Now the processed files keep theirs original name in and are put in an "esrgan" subfolder of the original file folder's.
-You can include the processing steps in the final filename (for exemple a file "test1" will be named "file1.(doublestep).(upresnet10-2-box).(modelSatsuki.v1.pth).(postfilter-0.5)"
-Added a link to the alsa64 esrgan model github to download more esrgan models.
sorry for the delay, I saw now that you released a new version, I will test it and if it gives the same error back ;)
-
I'm getting an error message that keeps saying the program can't find Python, even though I literally installed Python. Any ideas?
(https://i.imgur.com/y9dVfOa.jpg)
(https://i.imgur.com/vlcNLqG.jpg)
(https://i.imgur.com/uCrZUOv.jpg)
[/q
so I tested this new version and now gives this error: https://i.imgur.com/MwEFJwR.png
first type goes, but if I just change the model or put another image then this erro.
already if I try to make 1080p images of this error: https://i.imgur.com/gsThUqB.png
this is the setting I put: https://i.imgur.com/L9JPJiD.png
-
-The runtime error i'll check it
-Your computer can't process the prefilter of your 1080p source, try to disable prefilter and you'll see it's work
ATM prefilter don't split source so if you'r low on gpu ram, big pictures won't be processed, i need to found a way to split source file before prefiltering.
May take me some time to do it
-
It seems like splitting works but it's a gamble on if the joining actually aligns.
also there is an issue where if you run consecutive batches it gives an error and you need to reboot the program. annoying.
-
I'll check that asap
-
Version 1.1.3 released :
Changelog:
-Corrected the split source function again (must be ok now).
-Corrected a file accès bug if you change settings between 2 batchs
-Removed the finish popup at the end of processing, you can browse the log before closing it
-
Version 1.1.5 released :
Changelog:
-Corrected issue with esrgan script and unicode characters.
-Reducted the real time log to 500 lines to optimise speed.
-Added video processing (beta but no bug so far with my tests)
-
If esrganresizer doesn't support changing the output codec and output file extension for video. I would like to request that.
We think that FF8 remaster requires a specific codec for video. So would like to make the output video generated by this tool the correct format to save from converting it afterwards.
I figure you have ffmpeg for that. So if we can alter the commands passed to ffmpeg. Then all done. :)
Thank you.
-
Not planned ATM.
I use x264 as output as it's a standard (and aali driver can play it with the ffmpeg plugin it provides).
If you want to use specific codec just don't use "output to video" option and convert the png back to the video format you want, with ffmpeg or vdub2 ... or use x264 crf 0 (losseless) then transcode it to your favorite codec after
-
Are there any plans to update this tool? It does not currently work with many filters (like any 1x model, which is all the noise/artifact removal ones) because the math for tile sizes/overlaps is broken for anything but the default 4x scale.
This is unfortunate because many of esrgan's 1x models are amazing, like the JPEG ones and dedither, and the only way to use them is to manually edit the bash scripts every time you want to change between models with different scales.
It's kind of amazing, really, how esrgan's JPEG models completely blow away any of the expensive commercial denoiser tools, even those that focus on jpeg specifically.
-
You can use any 1x model
-disable prefilter if you don't need it
-set up "esrgan model" type to the model size (from 1x to 16x)
-
Hello and thank you for this tool. I have such error: 1 provonameforesragnprocess.0000000000000
ERROR: 1 provonameforesragnprocess.0000000000000
-
if you want to cut the source picture you can use this free tool:
https://colortarget.wordpress.com/pix-cut/
you will also be able to restitch the result once processed
-
Please use this instead : https://github.com/n00mkrad/cupscale (more up to date and easyier to use)