Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 09:14:35 +02:00
Code Issues Packages Projects Releases Wiki Activity

refactor: integrate sharpfin for high-quality image resize

Browse Source 
Vendor sharpfin library (Apache 2.0) and add centralized wrapper
module (images_sharpfin.py) replacing torchvision tensor/PIL
conversion and resize operations throughout the codebase.

- Add modules/sharpfin/ vendored library with MKS2021, Lanczos3,
  Mitchell, Catmull-Rom kernels and optional Triton sparse acceleration
- Add modules/images_sharpfin.py wrapper with to_tensor(), to_pil(),
  pil_to_tensor(), normalize(), resize(), resize_tensor()
- Add resize_quality and resize_linearize_srgb settings
- Add MKS2021 and Lanczos3 upscaler entries
- Replace torchvision.transforms.functional imports across 18 files
- to_pil() auto-detects HWC/BHWC layout, adds .round() before uint8
- Sparse Triton path falls back to dense GPU on compilation failure
- Mixed-axis resize splits into two single-axis scale() calls
- Masks and non-sRGB data always use linearize=False
This commit is contained in:
CalamitousFelicitousness
2026-02-10 00:13:35 +00:00
committed by vladmandic
parent 2c4d0751d9
commit 76aa949a26
30 changed files with 2878 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/pixelsmith_ext.py
+2 -6
View File
@@ -26,17 +26,13 @@ class Script(scripts_manager.Script):
def encode(self, p: processing.StableDiffusionProcessing, image: Image.Image):
if image is None:
return None
import numpy as np
import torch
from modules import images_sharpfin
if p.width is None or p.width == 0:
p.width = int(8 * (image.width * p.scale_by // 8))
if p.height is None or p.height == 0:
p.height = int(8 * (image.height * p.scale_by // 8))
image = images.resize_image(p.resize_mode, image, p.width, p.height, upscaler_name=p.resize_name, context=p.resize_context)
tensor = np.array(image).astype(np.float16) / 255.0
tensor = tensor[None].transpose(0, 3, 1, 2)
# image = image.transpose(0, 3, 1, 2)
tensor = torch.from_numpy(tensor).to(device=devices.device, dtype=devices.dtype)
tensor = images_sharpfin.to_tensor(image).unsqueeze(0).to(device=devices.device, dtype=devices.dtype)
tensor = 2.0 * tensor - 1.0
with devices.inference_context():
latent = shared.sd_model.vae.tiled_encode(tensor)