stable-diffusion-webui/modules/swinir.py

124 lines
4.0 KiB
Python
Raw Normal View History

import sys
import traceback
import cv2
import os
2022-09-20 11:09:13 -06:00
import contextlib
import numpy as np
from PIL import Image
import torch
import modules.images
from modules.shared import cmd_opts, opts, device
from modules.swinir_arch import SwinIR as net
2022-09-20 11:09:13 -06:00
precision_scope = (
torch.autocast if cmd_opts.precision == "autocast" else contextlib.nullcontext
)
def load_model(filename, scale=4):
model = net(
upscale=scale,
in_chans=3,
img_size=64,
window_size=8,
img_range=1.0,
depths=[6, 6, 6, 6, 6, 6, 6, 6, 6],
embed_dim=240,
num_heads=[8, 8, 8, 8, 8, 8, 8, 8, 8],
mlp_ratio=2,
upsampler="nearest+conv",
resi_connection="3conv",
)
pretrained_model = torch.load(filename)
2022-09-20 07:36:20 -06:00
model.load_state_dict(pretrained_model["params_ema"], strict=True)
2022-09-20 11:09:13 -06:00
if not cmd_opts.no_half:
model = model.half()
return model
def load_models(dirname):
for file in os.listdir(dirname):
path = os.path.join(dirname, file)
model_name, extension = os.path.splitext(file)
if extension != ".pt" and extension != ".pth":
continue
2022-09-20 11:09:13 -06:00
try:
modules.shared.sd_upscalers.append(UpscalerSwin(path, model_name))
except Exception:
print(f"Error loading SwinIR model: {path}", file=sys.stderr)
print(traceback.format_exc(), file=sys.stderr)
def upscale(
img,
model,
2022-09-20 16:02:56 -06:00
tile=opts.SWIN_tile,
tile_overlap=opts.SWIN_tile_overlap,
2022-09-20 11:09:13 -06:00
window_size=8,
scale=4,
):
2022-09-20 07:36:20 -06:00
img = np.array(img)
img = img[:, :, ::-1]
img = np.moveaxis(img, 2, 0) / 255
img = torch.from_numpy(img).float()
img = img.unsqueeze(0).to(device)
with torch.no_grad(), precision_scope("cuda"):
_, _, h_old, w_old = img.size()
h_pad = (h_old // window_size + 1) * window_size - h_old
w_pad = (w_old // window_size + 1) * window_size - w_old
2022-09-20 11:09:13 -06:00
img = torch.cat([img, torch.flip(img, [2])], 2)[:, :, : h_old + h_pad, :]
img = torch.cat([img, torch.flip(img, [3])], 3)[:, :, :, : w_old + w_pad]
output = inference(img, model, tile, tile_overlap, window_size, scale)
2022-09-20 11:09:13 -06:00
output = output[..., : h_old * scale, : w_old * scale]
output = output.data.squeeze().float().cpu().clamp_(0, 1).numpy()
if output.ndim == 3:
2022-09-20 11:09:13 -06:00
output = np.transpose(
output[[2, 1, 0], :, :], (1, 2, 0)
) # CHW-RGB to HCW-BGR
output = (output * 255.0).round().astype(np.uint8) # float32 to uint8
2022-09-20 11:09:13 -06:00
return Image.fromarray(output, "RGB")
def inference(img, model, tile, tile_overlap, window_size, scale):
# test the image tile by tile
b, c, h, w = img.size()
tile = min(tile, h, w)
assert tile % window_size == 0, "tile size should be a multiple of window_size"
sf = scale
stride = tile - tile_overlap
2022-09-20 11:09:13 -06:00
h_idx_list = list(range(0, h - tile, stride)) + [h - tile]
w_idx_list = list(range(0, w - tile, stride)) + [w - tile]
E = torch.zeros(b, c, h * sf, w * sf, dtype=torch.half, device=device).type_as(img)
W = torch.zeros_like(E, dtype=torch.half, device=device)
for h_idx in h_idx_list:
for w_idx in w_idx_list:
2022-09-20 11:09:13 -06:00
in_patch = img[..., h_idx : h_idx + tile, w_idx : w_idx + tile]
out_patch = model(in_patch)
out_patch_mask = torch.ones_like(out_patch)
2022-09-20 11:09:13 -06:00
E[
..., h_idx * sf : (h_idx + tile) * sf, w_idx * sf : (w_idx + tile) * sf
].add_(out_patch)
W[
..., h_idx * sf : (h_idx + tile) * sf, w_idx * sf : (w_idx + tile) * sf
].add_(out_patch_mask)
output = E.div_(W)
2022-09-20 07:36:20 -06:00
return output
2022-09-20 11:09:13 -06:00
2022-09-20 07:36:20 -06:00
class UpscalerSwin(modules.images.Upscaler):
2022-09-20 11:09:13 -06:00
def __init__(self, filename, title):
2022-09-20 07:36:20 -06:00
self.name = title
2022-09-20 11:09:13 -06:00
self.model = load_model(filename)
2022-09-20 07:36:20 -06:00
def do_upscale(self, img):
2022-09-20 11:09:13 -06:00
model = self.model.to(device)
img = upscale(img, model)
2022-09-20 07:36:20 -06:00
return img