ptrsuder / ieu.winforms Goto Github PK
View Code? Open in Web Editor NEWWindows only GUI for ESRGAN with additional features
License: GNU General Public License v3.0
Windows only GUI for ESRGAN with additional features
License: GNU General Public License v3.0
Describe the bug
When launching with these settings I get this error
[15/09/2022 11:36:43] Microsoft Windows 10.0.19043
[15/09/2022 11:36:43] .NET Framework 4.8.4515.0
[15/09/2022 11:36:43] NVIDIA GeForce RTX 3080 Ti: 12398 MB
[15/09/2022 11:36:43] Currently available VRAM: 581 MB
[15/09/2022 11:36:43] preview.png SPLIT START
[15/09/2022 11:36:44] preview.png SPLIT DONE
[15/09/2022 11:36:44] Starting ESRGAN...
[15/09/2022 11:36:45] Traceback (most recent call last):
[15/09/2022 11:36:45] File "upscale.py", line 12, in
[15/09/2022 11:36:45] upscaleSize = int(sys.argv[2])
[15/09/2022 11:36:45] ValueError: invalid literal for int() with base 10: '--input'
[15/09/2022 11:36:45] Error ocured during ESRGAN work!
To Reproduce
My settings:
Expected behavior
No error
Screenshots
Version
IEU Core V0.14.0
IEU Winforms 0.13.0
Additional context
Add any other context about the problem here.
Went form RTX 2080 TI to RTX 3090 on Windows 10. 8700K, 32GB RAM. On latest drivers 457.09 (tried 2 previous versions also).
Hitting Preview or "enhance and save as png" etc does nothing, the progress bar animates and python runs in the background but GPU never allocates any memory usage and the task never completes.
I've run into an issue where textures come out of the split brighter unless I split by RGB and then merge. Simply splitting is enough to produce the problem, but I assume it's related to splitting alphas. This doesn't affect dds or tiff files, but it does affect png files.
Is called crop-upscale-merge, should be IEU or Image Enhancement Utility.
Describe the bug
Whenever I try to use two models at the same time to upscale an image (so that I can easily compare the two results), the tiles do not merge. Instead, it prints out that it "attempted to divide by zero" and skips both results.
To Reproduce
Using these settings:
Expected behavior
The tiles should properly merge, resulting in two folders in the output folder, one for each model.
Screenshots
The error in question:
Version
Additional notes
I started having this problem as soon as I updated IEU. It's possible that the update messed something up somehow. In the meantime, I have to use each model one at a time to avoid the bug, and rename each image to differentiate between them.
Started happening lately. I have "auto set tille dimensions" enabled which sets tile res to 893x893 and it always fails to upscale,sometimes lowering it help but most of the time I still get the same error. I have RTX 2080 8GB.
[12.10.2020 14:30:26] Traceback (most recent call last): [12.10.2020 14:30:26] File "IEU_test.py", line 119, in <module> [12.10.2020 14:30:26] output = model(img_LR).data.squeeze(0).float().cpu().clamp_(0, 1).numpy() [12.10.2020 14:30:26] File "C:\Users\User\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\module.py", line 722, in _call_impl [12.10.2020 14:30:26] result = self.forward(*input, **kwargs) [12.10.2020 14:30:26] File "C:\ESRGAN\architecture.py", line 37, in forward [12.10.2020 14:30:26] x = self.model(x) [12.10.2020 14:30:26] File "C:\Users\User\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\module.py", line 722, in _call_impl [12.10.2020 14:30:26] result = self.forward(*input, **kwargs) [12.10.2020 14:30:26] File "C:\Users\User\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\container.py", line 117, in forward [12.10.2020 14:30:26] input = module(input) [12.10.2020 14:30:26] File "C:\Users\User\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\module.py", line 722, in _call_impl [12.10.2020 14:30:26] result = self.forward(*input, **kwargs) [12.10.2020 14:30:26] File "C:\Users\User\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\conv.py", line 419, in forward [12.10.2020 14:30:26] return self._conv_forward(input, self.weight) [12.10.2020 14:30:26] File "C:\Users\User\AppData\Local\Programs\Python\Python38\lib\site-packages\torch\nn\modules\conv.py", line 415, in _conv_forward [12.10.2020 14:30:26] return F.conv2d(input, weight, self.bias, self.stride, [12.10.2020 14:30:26] RuntimeError: CUDA out of memory. Tried to allocate 3.04 GiB (GPU 0; 8.00 GiB total capacity; 3.11 GiB already allocated; 2.77 GiB free; 3.13 GiB reserved in total by PyTorch) [12.10.2020 14:30:26] Error ocured during ESRGAN work!
Version: ImageEnhancingUtility.Winforms.0.13.0
When trying to merge tiles it takes more than 10 minutes to process each image (720x4000), when i let it process a couple of images it takes up enough ram to make my pc almost crash (30GB), it sometimes errors with (extract area etc)
To Reproduce
Expected behavior
merge tiles
Version
Seems like the app has problems with memory management. While doing the full process it crashes on merge. But if I open the app again and merge what's left in the folder it completes without problems. I've already turned the tile size down to 256px. It's impossible to do large texture folders.
IEU latest
RTX 2080S 8GB VRAM
16GB RAM
Win11
But I could also be blind, me missing things has happend often enough after all.
Not sure if ESRGAN has been updated to much since this came out but im curious if this will get updated to support it if it has?
I ran into a problem while running a batch of textures through a 1x denoise model. It only happened with one texture, although I don't see what's different about it. Doesn't happen anymore once I restart IEU and run "merge" again. Not sure what's going on or if you want to look into it, but here you go.
Files
LR.zip
I have a very heterogeneous set of images I want to run, resolution goes from 64x64 to 1024x1024, and I was wondering if there was a way to speed things up by packing smaller images together.
Say I allowed tiles up to 1000x1000 in the settings, and I want to process 10 images that are 500x500. I assume it would be possible to have IEU process 4 at a time instead of one after the other as it does now ? Would it actually be faster ?
Settings two different models and alpha 0.5, I also tried 0.3 but that didnt' work either.
Log:
[7/20/2019 11:33:35 AM] Alpha should be between 0.0 and 1.0
It would be cool if we could do this in image postprocess:
if >4096 than resize by 0.5
hi I am really really new to coding so I usually have no idea what's going on. I have run into an issue where they will have an error saying error module not found not module named 'architect' and i have no clue how to fix it after searching through online forums. PAny help will be appreciated thanksss
It's me again with another obscure problem I'm having. Tried a few different pictures, but I always get this error when I use a x1 model like denoise BC1. Works perfectly fine with x4 models.
Edit: Tested on another computer, I can't reproduce the problem there. I tried reinstalling IEU, ESRGAN and torch, no change.
The individual steps seem to work thou.
Describe the solution you'd like
A simple dark theme would be great
Describe the bug
While I can see that the program has successfully upscaled the tiles in the result folder, it is unable to merge said files because the "sequence contains no elements", even when just pressing the "Merge" button afterwards. I tried changing the filename of the original file (since it had +'s in it and I thought that might have caused issues), but that didn't seem to work. I also tried it with the "process with memory" option on, but not only did it not work, but the program also closed/crashed after throwing up the error message.
To Reproduce
What settings: See screenshots
Steps to reproduce the behavior:
Expected behavior
The program should merge the completed tiles after rendering them.
Version
[FEATURE] I would like IEU to be able to run inputs sequentially such as frames for a video, as of right now it sees the left most number and goes off of that which results in it going randomly down an input of frames. It would be very helpful to have the option to use the RIGHT most number to select order allowing it to run in numerical order.
Is your feature request related to a problem? Please describe.
The way I set up my python, the python installation with all the dependencies I need for ESRGAN always needs to be activated via "conda activate [environment name]" or, in a .bat script, like this: "call conda activate [environment name]". Without this, it will not work I think.
Describe the solution you'd like
Just a textbox to enter the conda environment name, if required, which is then called before the python script is executed so that the environment is correct.
Describe alternatives you've considered
None.
Additional context
That's it pretty much.
Good to see you're still developing this.
Can you please add an option to preserve folder structure while using multiple models chain?
It would be great to have a check all boxes option so we don't have to click each one when running multiple models.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.