GPEN-Windows
Cioscos opened this issue · comments
Hi!
I saw that you edited the code to work on Windows but from what I see it seems to be slower than "Linux" version.
Time ago, I created a stand-alone package to make GPEN upscaler run on Windows, with the original code. It doesn't requires python/MSVC/CUDA installed, just NVIDIA driver.
I would like to share this package here, maybe also in official README cause I'll upload it with any new commit.
That's awesome. What could I do for you?
That's awesome. What could I do for you?
I just wanted permission to share link. Currently my version is edited to work with DeepFaceLab images metadata which is a famous software to create deepfakes, but it works also with normal images. I can share last version in some minutes.
That's awesome. What could I do for you?
I just wanted permission to share link. Currently my version is edited to work with DeepFaceLab images metadata which is a famous software to create deepfakes, but it works also with normal images. I can share last version in some minutes.
Maybe you can share the link here, and I will add it to README.
That's awesome. What could I do for you?
I just wanted permission to share link. Currently my version is edited to work with DeepFaceLab images metadata which is a famous software to create deepfakes, but it works also with normal images. I can share last version in some minutes.
Maybe you can share the link here, and I will add it to README.
https://drive.google.com/file/d/1YJJVnPGq90e_mWZxSGGTptNQilZNfOEO/view?usp=drivesdk
This is the link. To make it work, you need to open install dependences bat file, first.
Gpen code is changed just for the scripts. You can use original gpen code and it will work.
Colorization works but gpen doesn't. It processes and outputs the file but it's exactly the same. Real Esrgan upscaling works too.
@echo off
call _internal\setenv.bat
"%PYTHON_EXECUTABLE%" face_enhancement.py ^
--model GPEN-BFR-512 ^
--use_sr ^
--indir examples/imgs ^
--outdir examples/outs
pause
It works like this though. With use_sr.
Colorization works but gpen doesn't. It processes and outputs the file but it's exactly the same. Real Esrgan upscaling works too.
@echo off call _internal\setenv.bat
"%PYTHON_EXECUTABLE%" face_enhancement.py ^ --model GPEN-BFR-512 ^ --use_sr ^ --indir examples/imgs ^ --outdir examples/outs
pause
It works like this though. With use_sr.
I updated the version in the link, now works fine.
Is there anything other than fixed bats? If yes can you just share them. I don't want to download the whole thing again.
Is there anything other than fixed bats? If yes can you just share them. I don't want to download the whole thing again.
I still didn't create GitHub repository, so or you wait for it, or you download again rar file. There are fixes in both bat files and python files.
Which link?
This is the old one.
It is always the same link. I update version. Use it, it will download last version.
Which link?
This is the old one.
From the same link I updated the package. Now you don't have to download everytime the entire pack.
@yangxy is there an idea about release date of training code?
The new code has been updated. Could you please help us to update it again? Thank you very much!
The new code has been updated. Could you please help us to update it again? Thank you very much!
No problem. Just use the link I shared or that you find in the GPEN readme.md in the main repository. This time will be the last which you will download the package 'cause this new version of my package can be updated using a bat file inside it.
The new code has been updated. Could you please help us to update it again? Thank you very much!
No problem. Just use the link I shared or that you find in the GPEN readme.md in the main repository. This time will be the last which you will download the package 'cause this new version of my package can be updated using a bat file inside it.
Important:
Some user reported that kaspersky antivirus detected a file of the package as virus put it is false positive. It is a file of git, cause my package has git installation inside.
Please send me an email. I can tell you how to train GPEN.
Please send me an email. I can tell you how to train GPEN.
Please send me an email. I can tell you how to train GPEN.
I will release the training codes. Stay tuned.
I keep getting this error:
E:\GPEN\GPEN-Windows_internal\Python37\python.exe: can't open file 'face_enhancement.py': [Errno 2] No such file or directory
can you help me?
I keep getting this error:
E:\GPEN\GPEN-Windows_internal\Python37\python.exe: can't open file 'face_enhancement.py': [Errno 2] No such file or directorycan you help me?
You need to first open misc) Install or update GPEN.bat
thank you :) Sorry for being boring. I was familiar with deepface and it was working on my comp without any problems (there was clear sequence when you staring the process, (finding faces, training... etc) I can't get this one)
now Im getting this message:
################ No landmarks in file input/imgs\70717071-660182505-397740380.jpg
0 70717071-660182505-397740380.jpg
Press any key to continue . . .
thank you :) Sorry for being boring. I was familiar with deepface and it was working on my comp without any problems (there was clear sequence when you staring the process, (finding faces, training... etc) I can't get this one)
now Im getting this message:
################ No landmarks in file input/imgs\70717071-660182505-397740380.jpg
0 70717071-660182505-397740380.jpg
Press any key to continue . . .
It is just a warning message. If the image processed is a normal image, that warning will be printed in the console, otherwise it means that the image gave in input is a DFL image, so it contains DFL metadata.
@yangxy Hi! Thank you very much for training code. I've just a question. As training dataset can we use a set without any landmarks or we need to find a specific type of set?
@Cioscos I downloaded and installed everything correctly but when I click any .bat file it says " all weights are already downloaded" and a line blinking below it. I wait for a few minutes, but nothing happens. Any ideas?
@Cioscos I downloaded and installed everything correctly but when I click any .bat file it says " all weights are already downloaded" and a line blinking below it. I wait for a few minutes, but nothing happens. Any ideas?
If you emailed me I answered there.
@Cioscos I downloaded and installed everything correctly but when I click any .bat file it says " all weights are already downloaded" and a line blinking below it. I wait for a few minutes, but nothing happens. Any ideas?
If you emailed me I answered there.
I waited 5+ minutes but still nothing happens.
@Cioscos I downloaded and installed everything correctly but when I click any .bat file it says " all weights are already downloaded" and a line blinking below it. I wait for a few minutes, but nothing happens. Any ideas?
If you emailed me I answered there.
I waited 5+ minutes but still nothing happens.
Open tasks manager and after bat file and see what it happens under cmd voice.
@Cioscos I downloaded and installed everything correctly but when I click any .bat file it says " all weights are already downloaded" and a line blinking below it. I wait for a few minutes, but nothing happens. Any ideas?
If you emailed me I answered there.
I waited 5+ minutes but still nothing happens.
Open tasks manager and after bat file and see what it happens under cmd voice.
Nothing was happening. I had to run “clear_local_env.bat” from _internal and everything’s fine now. Thanks!
Hello. Does CPU version still requires CUDA videocard? Because i get this message:
Found GPU%d %s which is of cuda capability %d.%d. PyTorch no longer supports this GPU because it is too old. The minimum cuda capability supported by this library is %d.%d.
ValueError: Unknown CUDA arch (3.0+PTX) or GPU not supported
they all dropped cuda compute 3.5 cards , lower than compute 5 wont work, sucks big time , had to buy new card.
They skimp on space so they wont even compile one pytorch cuda special version every half year to support older gpus and yet they keep releasing new versions that take up lots of GB of space all the time with minor differences... go figure.
Yeah in one year maybe even compute 5 wont be supported, this is just plain wrong.