Merge pull request #14524 from akx/fix-swinir-issues

Fix SwinIR issues

extensions-builtin/SwinIR/scripts/swinir_model.py

 import logging
 import sys
 
+import torch
 from PIL import Image
 
 from modules import devices, modelloader, script_callbacks, shared, upscaler_utils
@@ -50,7 +51,7 @@ class UpscalerSwinIR(Upscaler):
             model,
             tile_size=shared.opts.SWIN_tile,
             tile_overlap=shared.opts.SWIN_tile_overlap,
-            scale=4,  # TODO: This was hard-coded before too...
+            scale=model.scale,
             desc="SwinIR",
         )
         devices.torch_gc()
@@ -69,7 +70,7 @@ class UpscalerSwinIR(Upscaler):
         model_descriptor = modelloader.load_spandrel_model(
             filename,
             device=self._get_device(),
-            dtype=devices.dtype,
+            prefer_half=(devices.dtype == torch.float16),
             expected_architecture="SwinIR",
         )
         if getattr(shared.opts, 'SWIN_torch_compile', False):

modules/upscaler_utils.py

@@ -94,6 +94,7 @@ def tiled_upscale_2(
     tile_size: int,
     tile_overlap: int,
     scale: int,
+    device: torch.device,
     desc="Tiled upscale",
 ):
     # Alternative implementation of `upscale_with_model` originally used by
@@ -101,9 +102,6 @@ def tiled_upscale_2(
     # weighting is done in PyTorch space, as opposed to `images.Grid` doing it in
     # Pillow space without weighting.
 
-    # Grab the device the model is on, and use it.
-    device = torch_utils.get_param(model).device
-
     b, c, h, w = img.size()
     tile_size = min(tile_size, h, w)
 
@@ -175,7 +173,8 @@ def upscale_2(
     """
     Convenience wrapper around `tiled_upscale_2` that handles PIL images.
     """
-    tensor = pil_image_to_torch_bgr(img).float().unsqueeze(0)  # add batch dimension
+    param = torch_utils.get_param(model)
+    tensor = pil_image_to_torch_bgr(img).to(dtype=param.dtype).unsqueeze(0)  # add batch dimension
 
     with torch.no_grad():
         output = tiled_upscale_2(
@@ -185,5 +184,6 @@ def upscale_2(
             tile_overlap=tile_overlap,
             scale=scale,
             desc=desc,
+            device=param.device,
         )
     return torch_bgr_to_pil_image(output)