Upload folder using huggingface_hub

external/midas/__init__.py

@@ -1,20 +1,23 @@
 import cv2
 import numpy as np
 import torch
-
 from einops import rearrange
+
 from .api import MiDaSInference
 
-model = MiDaSInference(model_type="dpt_hybrid").cuda()
+model = None
 
 
 def apply_midas(input_image, a=np.pi * 2.0, bg_th=0.1):
+    global model
+    if not model:
+        model = MiDaSInference(model_type="dpt_hybrid").cuda()
     assert input_image.ndim == 3
     image_depth = input_image
     with torch.no_grad():
         image_depth = torch.from_numpy(image_depth).float().cuda()
         image_depth = image_depth / 127.5 - 1.0
-        image_depth = rearrange(image_depth, 'h w c -> 1 c h w')
+        image_depth = rearrange(image_depth, "h w c -> 1 c h w")
         depth = model(image_depth)[0]
 
         depth_pt = depth.clone()
@@ -30,7 +33,7 @@ def apply_midas(input_image, a=np.pi * 2.0, bg_th=0.1):
         x[depth_pt < bg_th] = 0
         y[depth_pt < bg_th] = 0
         normal = np.stack([x, y, z], axis=2)
-        normal /= np.sum(normal ** 2.0, axis=2, keepdims=True) ** 0.5
+        normal /= np.sum(normal**2.0, axis=2, keepdims=True) ** 0.5
         normal_image = (normal * 127.5 + 127.5).clip(0, 255).astype(np.uint8)
 
         return depth_image, normal_image

internals/pipelines/replace_background.py

@@ -172,14 +172,5 @@ class ReplaceBackground(AbstractPipeline):
         if not has_nsfw:
             for i in range(len(images)):
                 images[i].paste(image, (0, 0), image)
-                w, h = images[i].size
-                out_bytes = self.upscaler.upscale(
-                    image=images[i],
-                    width=w,
-                    height=h,
-                    face_enhance=False,
-                    resize_dimension=resize_dimension,
-                )
-                images[i] = Image.open(BytesIO(out_bytes)).convert("RGB")
 
         return (images, has_nsfw)

internals/pipelines/upscaler.py

@@ -127,7 +127,7 @@ class Upscaler:
             image = download_image(image)
 
         w, h = image.size
-        if max(w, h) > 1536:
+        if max(w, h) > 1024:
             image = ImageUtil.resize_image(image, dimension=1024)
 
         in_path = str(Path.home() / ".cache" / "input_upscale.png")