update

app.py

@@ -1,6 +1,6 @@
 import gradio as gr
 from loadimg import load_img
-import spaces
+#import spaces
 from transformers import AutoModelForImageSegmentation
 import torch
 from torchvision import transforms
@@ -11,10 +11,11 @@ import numpy as np
 import os
 import tempfile
 import uuid
+import devicetorch
 
 torch.set_float32_matmul_precision("medium")
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
+device = devicetorch.get(torch)
 
 birefnet = AutoModelForImageSegmentation.from_pretrained(
     "ZhengPeng7/BiRefNet", trust_remote_code=True
@@ -29,7 +30,7 @@ transform_image = transforms.Compose(
 )
 
 
-@spaces.GPU
+#@spaces.GPU
 def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=0, video_handling="slow_down"):
     try:
         # Load the video using moviepy
@@ -111,7 +112,7 @@ def fn(vid, bg_type="Color", bg_image=None, bg_video=None, color="#00FF00", fps=
 
 def process(image, bg):
     image_size = image.size
-    input_images = transform_image(image).unsqueeze(0).to("cuda")
+    input_images = transform_image(image).unsqueeze(0).to(device)
     # Prediction
     with torch.no_grad():
         preds = birefnet(input_images)[-1].sigmoid().cpu()
@@ -191,4 +192,4 @@ with gr.Blocks(theme=gr.themes.Ocean()) as demo:
     )
 
 if __name__ == "__main__":
-    demo.launch(show_error=True)
+    demo.launch(show_error=True)

requirements.txt

@@ -1,7 +1,6 @@
 torch
 accelerate
 opencv-python
-spaces
 pillow
 numpy
 timm
@@ -15,4 +14,5 @@ gradio
 gradio_imageslider
 loadimg>=0.1.1
 moviepy 
-pydub
+pydub
+devicetorch