2022-09-19 20:05:12 +00:00
|
|
|
import sys
|
|
|
|
import traceback
|
|
|
|
import cv2
|
|
|
|
import os
|
2022-09-20 17:09:13 +00:00
|
|
|
import contextlib
|
2022-09-19 20:05:12 +00:00
|
|
|
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 17:09:13 +00:00
|
|
|
|
|
|
|
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 13:36:20 +00:00
|
|
|
model.load_state_dict(pretrained_model["params_ema"], strict=True)
|
2022-09-20 17:09:13 +00:00
|
|
|
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
|
2022-09-19 20:05:12 +00:00
|
|
|
|
2022-09-20 17:09:13 +00:00
|
|
|
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,
|
|
|
|
tile=opts.GAN_tile,
|
|
|
|
tile_overlap=opts.GAN_tile_overlap,
|
|
|
|
window_size=8,
|
|
|
|
scale=4,
|
|
|
|
):
|
2022-09-20 13:36:20 +00:00
|
|
|
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)
|
2022-09-19 20:05:12 +00:00
|
|
|
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 17:09:13 +00:00
|
|
|
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]
|
2022-09-19 20:05:12 +00:00
|
|
|
output = inference(img, model, tile, tile_overlap, window_size, scale)
|
2022-09-20 17:09:13 +00:00
|
|
|
output = output[..., : h_old * scale, : w_old * scale]
|
2022-09-19 20:05:12 +00:00
|
|
|
output = output.data.squeeze().float().cpu().clamp_(0, 1).numpy()
|
|
|
|
if output.ndim == 3:
|
2022-09-20 17:09:13 +00:00
|
|
|
output = np.transpose(
|
|
|
|
output[[2, 1, 0], :, :], (1, 2, 0)
|
|
|
|
) # CHW-RGB to HCW-BGR
|
2022-09-19 20:05:12 +00:00
|
|
|
output = (output * 255.0).round().astype(np.uint8) # float32 to uint8
|
2022-09-20 17:09:13 +00:00
|
|
|
return Image.fromarray(output, "RGB")
|
|
|
|
|
|
|
|
|
2022-09-19 20:05:12 +00:00
|
|
|
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 17:09:13 +00:00
|
|
|
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)
|
2022-09-19 20:05:12 +00:00
|
|
|
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 17:09:13 +00:00
|
|
|
in_patch = img[..., h_idx : h_idx + tile, w_idx : w_idx + tile]
|
2022-09-19 20:05:12 +00:00
|
|
|
out_patch = model(in_patch)
|
|
|
|
out_patch_mask = torch.ones_like(out_patch)
|
|
|
|
|
2022-09-20 17:09:13 +00:00
|
|
|
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)
|
2022-09-19 20:05:12 +00:00
|
|
|
output = E.div_(W)
|
|
|
|
|
2022-09-20 13:36:20 +00:00
|
|
|
return output
|
2022-09-20 17:09:13 +00:00
|
|
|
|
|
|
|
|
2022-09-20 13:36:20 +00:00
|
|
|
class UpscalerSwin(modules.images.Upscaler):
|
2022-09-20 17:09:13 +00:00
|
|
|
def __init__(self, filename, title):
|
2022-09-20 13:36:20 +00:00
|
|
|
self.name = title
|
2022-09-20 17:09:13 +00:00
|
|
|
self.model = load_model(filename)
|
2022-09-20 13:36:20 +00:00
|
|
|
|
|
|
|
def do_upscale(self, img):
|
2022-09-20 17:09:13 +00:00
|
|
|
model = self.model.to(device)
|
|
|
|
img = upscale(img, model)
|
2022-09-20 13:36:20 +00:00
|
|
|
return img
|