sphinxsolution commited on
Commit
e956b51
·
verified ·
1 Parent(s): 786b29c

updated app.py

Browse files
Files changed (1) hide show
  1. app.py +45 -13
app.py CHANGED
@@ -1,22 +1,54 @@
 
 
1
  import gradio as gr
2
  import torch
3
- import cv2
4
  import numpy as np
5
- from lama_inpaint import Inpainter # Assuming you have a module for LaMa inpainting
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
 
7
- def inpaint_image(image, mask):
8
- inpainter = Inpainter()
9
- output = inpainter.inpaint(image, mask)
10
- return output
11
 
12
- description = "Upload an image and a mask (black for areas to keep, white for areas to inpaint)."
 
 
 
 
 
13
 
 
14
  demo = gr.Interface(
15
- fn=inpaint_image,
16
- inputs=[gr.Image(type='numpy'), gr.Image(type='numpy', label='Mask')],
17
- outputs=gr.Image(type='numpy'),
18
- title="Inpaint Anything with LaMa",
19
- description=description
20
  )
21
 
22
- demo.launch(share=True)
 
1
+ import os
2
+ import subprocess
3
  import gradio as gr
4
  import torch
 
5
  import numpy as np
6
+ import cv2
7
+
8
+ # Clone LaMa repo if not present
9
+ LAMA_PATH = "lama"
10
+ if not os.path.exists(LAMA_PATH):
11
+ subprocess.run(["git", "clone", "https://github.com/saic-mdal/lama.git", LAMA_PATH])
12
+ subprocess.run(["pip", "install", "-r", f"{LAMA_PATH}/requirements.txt"])
13
+
14
+ # Add LaMa to Python path
15
+ import sys
16
+ sys.path.append(LAMA_PATH)
17
+
18
+ # Import LaMa inpainting module
19
+ from lama.saicinpainting.evaluation.data import move_to_device
20
+ from lama.saicinpainting.training.trainers import load_checkpoint
21
+
22
+ # Load LaMa model
23
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
24
+ checkpoint_path = os.path.join(LAMA_PATH, "big-lama")
25
+
26
+ if not os.path.exists(checkpoint_path):
27
+ subprocess.run(["wget", "https://huggingface.co/spaces/pg56714/Inpaint-Anything/resolve/main/lama/big-lama.zip", "-O", "lama/big-lama.zip"])
28
+ subprocess.run(["unzip", "-o", "lama/big-lama.zip", "-d", "lama"])
29
+
30
+ model = load_checkpoint(checkpoint_path, map_location=device)
31
+ model = model.to(device).eval()
32
 
33
+ def inpaint(image, mask):
34
+ # Convert image and mask to proper tensor format
35
+ image = torch.tensor(image).permute(2, 0, 1).unsqueeze(0).float().to(device)
36
+ mask = torch.tensor(mask).unsqueeze(0).unsqueeze(0).float().to(device)
37
 
38
+ # Perform inpainting
39
+ with torch.no_grad():
40
+ inpainted = model(image, mask)
41
+
42
+ inpainted = inpainted.squeeze(0).permute(1, 2, 0).cpu().numpy()
43
+ return (inpainted * 255).astype(np.uint8)
44
 
45
+ # Create Gradio interface
46
  demo = gr.Interface(
47
+ fn=inpaint,
48
+ inputs=[gr.Image(type="numpy"), gr.Image(type="numpy", label="Mask")],
49
+ outputs=gr.Image(type="numpy"),
50
+ title="LaMa Inpainting",
51
+ description="Upload an image and a mask (white for areas to be inpainted)."
52
  )
53
 
54
+ demo.launch()