v4.1.1

[C0untFloyd]

Bugfix: Post-processing after swapping

---

This discussion was created from the release v4.1.1.

[RoopUser]

Thank you for everything! You are pioneering a whole new world of por... media! ;)

Question: Do I have to re-make my previous face-sets to get things working with the higher resolution swapping?

[RoopUser]

Sorry, should have tried... the answer seems to be no. But damn does it drag!
Normally with default settings at 128x I would be doing about 14fps and at 512x it dropped to about 1.5fps!

[etruscanmuffin909]

Question: Do I have to re-make my previous face-sets to get things working with the higher resolution swapping?

Can anyone confirm? I see that the results are distinctly better at 512 using an old fsz but does remaking the fsz lead to higher resolution results?

Thank you for this huge improvement

[C0untFloyd]

Normally with default settings at 128x I would be doing about 14fps and at 512x it dropped to about 1.5fps!

Yeah it's 4 times the resolution and 4 times the swapping process. But I believe it's worth it. Increase the swapping steps to 2 or 3 and the results are incredible, even without using any enhancer.

I see that the results are distinctly better at 512 using an old fsz but does remaking the fsz lead to higher resolution results?

No need to redo them if they were already at a good enough resolution.

[hex17or]

Thanks, eager to try out this build. On a slightly different note, any chance you know of any other repo or method that I can use to convert my xseg model to onnx? The one you linked in v3.9.0 release notes here seems to not be available anymore.

[C0untFloyd]

Oooh damn! The origin and the fork both disappeared and there doesn't seem to be an archived version. Weird...
Here's my last pulled version
XSegNet2onnx.zip
Perhaps you can create a repo as backup out of this.

[seneka11]

Hello. Tell me, what am I doing wrong? The installation was carried out as in the YouTube video. At the end of the installation I received the following message:
(venv) D:\roop-unleashed-main>python run.py
Using provider [] - Device:cpu
D:\roop-unleashed-main\venv\lib\site-packages\gradio\components\dropdown.py:187: UserWarning: The value passed into gr.Dropdown() is not in the list of choices. Please update the list of choices to include: cuda or set allow_custom_value=True.
warnings.warn(

In the roop tab in settings there is no option to select cuda
In the requirements file I have the following parameters:
numpy==1.26.4
gradio==4.38.1
opencv-python==4.9.0.80
onnx==1.16.0
insightface==0.7.3
psutil==5.9.6
onnxruntime==1.15.1
onnxruntime-gpu==1.15.1
tqdm==4.66.4
ftfy
regex
pyvirtualcam

[seneka11]

I think this is related to the installation of the requirements file. If you copy and install the old requirements file with parameters:

--extra-index-url https://download.pytorch.org/whl/cu118

numpy==1.26.4
gradio==4.38.1
opencv-python==4.9.0.80
onnx==1.16.0
insightface==0.7.3
psutil==5.9.6
torch==2.1.2+cu118; sys_platform != 'darwin'
torch==2.1.2; sys_platform == 'darwin'
torchvision==0.16.2+cu118; sys_platform != 'darwin'
torchvision==0.16.2; sys_platform == 'darwin'
onnxruntime==1.17.1; sys_platform == 'darwin' and platform_machine != 'arm64'
onnxruntime-silicon==1.16.3; sys_platform == 'darwin' and platform_machine == 'arm64'
onnxruntime-gpu==1.17.1; sys_platform != 'darwin'
tqdm==4.66.4
ftfy
regex
pyvirtualcam

Then cuda is available in the settings and it works but with " None Select post-processing" . If I select, for example, "GPEN" or any other, I get an error.

[C0untFloyd]

as in the YouTube video

Huh? There is no youtube video known to me? If you're on windows just use the 1-click installer and of course there's the wiki

[seneka11]

Thank you.

[seneka11]

Problem solved. I used the automatic installer. If anyone encounters the problem "Conda not installed" during installation ( #346 )
you need to delete all unnecessary paths to roop-unleashed in "Changing Environment Variables".

[seneka11]

When i choose Select post-processing, for example 'GPEN ' I get an error
onnxruntime.capi.onnxruntime_pybind11_state.RuntimeException: [ONNXRuntimeError] : 6 : RUNTIME_EXCEPTION : Exception during initialization: D:\a_work\1\s\onnxruntime\core\framework\bfc_arena.cc:376 onnxruntime::BFCArena::AllocateRawInternal Failed to allocate memory for requested buffer of size 18874368

Tell me please how I can fix this =( I use cuda Provider

[C0untFloyd]

Why did you install a 2nd time if it was working? Anyway, how much VRAM do you have? Did you perhaps use CPU instead of CUDA the first time?

[seneka11]

gtx 1060 3 gb . No.. only CUDA . Maybe I'll reinstall the system.. Most likely the installed files conflict with each other and I need to do proper cleaning, but I don't know how to do this. It's easier to reinstall the entire system.

[seneka11]

Because this is not the only such error. There are other errors not related to memory.

[C0untFloyd]

I'm sorry to break it to you but 3 Gb really is not enough to use this. In my experience 6 Gb and a GPU in the ranks of a 1080 GT is the bare minimum and then you probably won't be able to swap 4k videos and use post-processors. You really should upgrade your gpu and until then just use cpu as provider.

[seneka11]

Yes. You are right. I'm getting memory errors unfortunately. Thanks for the help !!

[Bakulat]

Excited about the 512! I haven't run roop in a couple weeks, updated it at launch and just started using it as normal - noticed when I had post-processing it has slowed to a crawl. getting through 200 frames of 720p is now impossible where it took a 2-3 minutes before. I kept it at 128 since I hadn't even noticed that upgrade yet. So now I'm playing around with using 512 instead of post processing, but so far I think I might prefer GFPGAN / GPEN, GFP I think it was all of a sudden was working incredibly well a few weeks ago, like stunning results. I've gotten so much fun and use out of this, so I would like to thank you dearly! Just reporting my experience, and wondering what might be the cause of the sudden slowdown. Using cuda, and a 2060

[C0untFloyd]

As I wrote previously the artificial upscaling is a sort of tricky upscaling where the swapping process is repeated multiple times and the results are patched together for a higher resolution. With 512px this whole process runs 4 times and is at least 4 times slower. Leaving it at 128px shouldn't be a noticeable difference to previous versions.

[Bakulat]

I ended up going back to 4.0, and then 3.9, and I'm still at a crawl. It may be something on my end?

…

On Sat, Jul 27, 2024 at 4:52 PM C0untFloyd ***@***.***> wrote: As I wrote previously the artificial upscaling is a sort of tricky upscaling where the swapping process is repeated multiple times and the results are patched together for a higher resolution. With 512px this whole process runs 4 times and is at least 4 times slower. Leaving it at 128px shouldn't be a noticeable difference to previous versions. — Reply to this email directly, view it on GitHub <#805 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BJSP7MJKQARAR6HGERJHCDDZOQB7PAVCNFSM6AAAAABK446YGKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTAMJWHEYDEMY> . You are receiving this because you commented.Message ID: ***@***.*** com>

[Bakulat]

For me its directly related to the post-processing even at 3.9. Moves super fast until post-processing is turned on at all, even at 3.9

…

On Sat, Jul 27, 2024 at 9:41 PM Larry Jeezer ***@***.***> wrote: I ended up going back to 4.0, and then 3.9, and I'm still at a crawl. It may be something on my end? On Sat, Jul 27, 2024 at 4:52 PM C0untFloyd ***@***.***> wrote: > As I wrote previously the artificial upscaling is a sort of tricky > upscaling where the swapping process is repeated multiple times and the > results are patched together for a higher resolution. With 512px this whole > process runs 4 times and is at least 4 times slower. Leaving it at 128px > shouldn't be a noticeable difference to previous versions. > > — > Reply to this email directly, view it on GitHub > <#805 (reply in thread)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/BJSP7MJKQARAR6HGERJHCDDZOQB7PAVCNFSM6AAAAABK446YGKVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTAMJWHEYDEMY> > . > You are receiving this because you commented.Message ID: > ***@***.*** > com> >