-
Notifications
You must be signed in to change notification settings - Fork 27k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Use Spandrel for upscaling and face restoration architectures #14425
Conversation
b1a61e9
to
e61c70b
Compare
ecee8df
to
8bfe7cf
Compare
@@ -185,8 +180,7 @@ def on_ui_settings(): | |||
|
|||
shared.opts.add_option("SWIN_tile", shared.OptionInfo(192, "Tile size for all SwinIR.", gr.Slider, {"minimum": 16, "maximum": 512, "step": 16}, section=('upscaling', "Upscaling"))) | |||
shared.opts.add_option("SWIN_tile_overlap", shared.OptionInfo(8, "Tile overlap, in pixels for SwinIR. Low values = visible seam.", gr.Slider, {"minimum": 0, "maximum": 48, "step": 1}, section=('upscaling', "Upscaling"))) | |||
if int(torch.__version__.split('.')[0]) >= 2 and platform.system() != "Windows": # torch.compile() require pytorch 2.0 or above, and not on Windows |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We're always torch >= 2.0, and we now just try compile
without checking the platform.
.github/workflows/run_tests.yaml
Outdated
env: | ||
IGNORE_CMD_ARGS_ERRORS: "1" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The new test will fail without this since a module attempts to read the pytest
options as regular webui arguments.
img, | ||
tile_size=opts.ESRGAN_tile, | ||
tile_overlap=opts.ESRGAN_tile_overlap, | ||
# TODO: `outscale`? |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Do we need to downscale too-large images here according to info.scale
? AIUI, there might be some other process that also does that?
modules/upscaler_utils.py
Outdated
return Image.fromarray(output, 'RGB') | ||
|
||
|
||
def upscale_with_model(model, img: Image.Image, *, tile_size: int, tile_overlap: int = 0): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is used by both esrgan
and realesrgan
.
def inference( | ||
img, | ||
model, | ||
*, | ||
tile: int, | ||
tile_overlap: int, | ||
window_size: int, | ||
scale: int, | ||
device, | ||
): |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This smells like tiled_upscale
, but with tile overlap handled by weight scaling.
ce58f5d
to
37458e6
Compare
Oh yeah, I've been using this PR for a couple days now; it works. |
@gel-crabs Thanks for trying it out! I (force-)pushed this branch to update spandrel to a newer version, as well as add experimental support for HAT upscalers, if you want to try that out. (You'll need to bring your own models and put them in |
It works! Admittedly it has issues with deepcache where it adds black splotches to the image during hires fix, but otherwise working. I tried to hack in support for DAT as well by copying hat_model.py and replacing HAT with DAT, but it just made the image go full black. Edit: It actually has nothing to do with deepcache, or any extensions at all. I'm going to try testing with different models. I tried with a different 4x HAT upscaler and it gives full black images, so the HAT support doesn't seem to be working correctly. |
I'm generally not pumped about adding new dependencies, but this removes a lot of code we just copy pasted, so that seems nice. Some questions:
|
I think this actually leads to less dependencies in total (I'll run the numbers later). The Spandrel folks seem nice and responsive too. :)
Autogenerated by PyCharm when refactoring code. Will yeet, my bad.
Also accidentally added to this PR (since I was tired of having a gazillion WebUI tabs get auto-opened), my bad. Will yeet.
I can also add an
Since we do I'll add a simple "output image was different" check!
Good question - since Spandrel auto-detects the model arch from the checkpoint, it'd happily load it, and maybe fail with a parameter error down the line when we try to call the architecture with kwargs it doesn't get. I can add
I did, on my machine (Macbook). |
Looks like SwinIR x2 is not working now. I get this in any model:
|
@wcde Thanks, I'll take a peek – what's your SwinIR tile size and overlap setting, and the size of the image you're trying to upscale? |
In code hardcoded scale to 4.
Second problem - model is loaded with dtype
Which give:
|
@wcde In fairness,
I'll take a look at the |
I guess it will happen with a lot of extensions after updating. Maybe it should be mentioned in changelog? |
Description
This PR yeets most of the copy-pasted or otherwise vendored model architectures in favor of just using
Spandrel.
Converted models are:
Not converted is LDSR; it doesn't exist in Spandrel.
There's still some more cleanup that could be done – there are multiple implementations of tiled inference right now, for one, and the model loading/downloading/... code is kind of a mess (should continue where I left off with Upscaler model loading cleanup #10823), but I'll hold off on that for this PR.
As an added bonus, this adds (experimental, works-on-my-machine) support for HAT upscaling models.
Screenshots/videos:
No visual changes. This seems to Work On My Machine but it'd be lovely if someone else tried this out too.
Checklist: