Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,138 +1,100 @@
|
|
1 |
import os
|
2 |
-
os.system("wget https://huggingface.co/Carve/LaMa-ONNX/resolve/main/lama_fp32.onnx")
|
3 |
-
os.system("pip install onnxruntime imageio")
|
4 |
-
import cv2
|
5 |
-
import paddlehub as hub
|
6 |
-
import gradio as gr
|
7 |
-
import torch
|
8 |
-
from PIL import Image, ImageOps
|
9 |
-
import numpy as np
|
10 |
import imageio
|
11 |
-
|
12 |
-
|
13 |
-
model = hub.Module(name='U2Net')
|
14 |
import cv2
|
15 |
-
import
|
16 |
import onnxruntime
|
17 |
-
import torch
|
18 |
-
from PIL import Image
|
19 |
-
sess_options = onnxruntime.SessionOptions()
|
20 |
-
rmodel = onnxruntime.InferenceSession('lama_fp32.onnx', sess_options=sess_options)
|
21 |
-
|
22 |
-
# Source https://github.com/advimman/lama
|
23 |
-
def get_image(image):
|
24 |
-
if isinstance(image, Image.Image):
|
25 |
-
img = np.array(image)
|
26 |
-
elif isinstance(image, np.ndarray):
|
27 |
-
img = image.copy()
|
28 |
-
else:
|
29 |
-
raise Exception("Input image should be either PIL Image or numpy array!")
|
30 |
-
|
31 |
-
if img.ndim == 3:
|
32 |
-
img = np.transpose(img, (2, 0, 1)) # chw
|
33 |
-
elif img.ndim == 2:
|
34 |
-
img = img[np.newaxis, ...]
|
35 |
-
|
36 |
-
assert img.ndim == 3
|
37 |
-
|
38 |
-
img = img.astype(np.float32) / 255
|
39 |
-
return img
|
40 |
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
return (x // mod + 1) * mod
|
46 |
|
|
|
|
|
47 |
|
48 |
-
|
49 |
-
if img.shape[0] == 1:
|
50 |
-
img = img[0]
|
51 |
-
else:
|
52 |
-
img = np.transpose(img, (1, 2, 0))
|
53 |
|
54 |
-
|
55 |
-
|
56 |
-
if
|
57 |
-
|
|
|
|
|
|
|
58 |
else:
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
return np.
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
def predict(jpg, msk):
|
95 |
-
|
96 |
-
|
97 |
-
imagex = Image.open(jpg)
|
98 |
-
mask = Image.open(msk).convert("L")
|
99 |
-
|
100 |
-
image, mask = prepare_img_and_mask(imagex.resize((512, 512)), mask.resize((512, 512)), 'cpu')
|
101 |
-
# Run the model
|
102 |
-
outputs = rmodel.run(None, {'image': image.numpy().astype(np.float32), 'mask': mask.numpy().astype(np.float32)})
|
103 |
-
|
104 |
output = outputs[0][0]
|
105 |
-
# Postprocess the outputs
|
106 |
output = output.transpose(1, 2, 0)
|
107 |
-
output = output.astype(np.uint8)
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
title
|
136 |
-
description
|
137 |
-
|
138 |
-
|
|
|
|
1 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
import imageio
|
3 |
+
from PIL import Image
|
4 |
+
import gradio as gr
|
|
|
5 |
import cv2
|
6 |
+
import paddlehub as hub
|
7 |
import onnxruntime
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
+
# Download and setup models
|
10 |
+
os.system("wget https://huggingface.co/Carve/LaMa-ONNX/resolve/main/lama_fp32.onnx")
|
11 |
+
os.system("pip install onnxruntime imageio")
|
12 |
+
os.makedirs("data", exist_ok=True)
|
13 |
+
os.makedirs("dataout", exist_ok=True)
|
14 |
|
15 |
+
# Load LaMa ONNX model
|
16 |
+
sess_options = onnxruntime.SessionOptions()
|
17 |
+
lama_model = onnxruntime.InferenceSession('lama_fp32.onnx', sess_options=sess_options)
|
|
|
18 |
|
19 |
+
# Load U^2-Net model for automatic masking
|
20 |
+
u2net_model = hub.Module(name='U2Net')
|
21 |
|
22 |
+
# --- Helper Functions ---
|
|
|
|
|
|
|
|
|
23 |
|
24 |
+
def prepare_image(image, target_size=(512, 512)):
|
25 |
+
"""Resizes and preprocesses image for LaMa model."""
|
26 |
+
if isinstance(image, Image.Image):
|
27 |
+
image = image.resize(target_size)
|
28 |
+
image = np.array(image)
|
29 |
+
elif isinstance(image, np.ndarray):
|
30 |
+
image = cv2.resize(image, target_size)
|
31 |
else:
|
32 |
+
raise ValueError("Input image should be either PIL Image or numpy array!")
|
33 |
+
|
34 |
+
# Normalize to [0, 1] and convert to CHW format
|
35 |
+
image = image.astype(np.float32) / 255.0
|
36 |
+
if image.ndim == 3:
|
37 |
+
image = np.transpose(image, (2, 0, 1))
|
38 |
+
elif image.ndim == 2:
|
39 |
+
image = image[np.newaxis, ...]
|
40 |
+
return image[np.newaxis, ...] # Add batch dimension
|
41 |
+
|
42 |
+
def generate_mask(image, method="automatic"):
|
43 |
+
"""Generates mask from image using U^2-Net or user input."""
|
44 |
+
if method == "automatic":
|
45 |
+
input_size = 320 # Adjust based on U^2-Net requirements
|
46 |
+
result = u2net_model.Segmentation(
|
47 |
+
images=[cv2.cvtColor(image, cv2.COLOR_RGB2BGR)],
|
48 |
+
paths=None,
|
49 |
+
batch_size=1,
|
50 |
+
input_size=input_size,
|
51 |
+
output_dir='output',
|
52 |
+
visualization=False
|
53 |
+
)
|
54 |
+
mask = Image.fromarray(result[0]['mask'])
|
55 |
+
mask = mask.resize((512, 512)) # Resize to match LaMa input
|
56 |
+
mask.save("./data/data_mask.png")
|
57 |
+
else: # "manual"
|
58 |
+
mask = imageio.imread("./data/data_mask.png")
|
59 |
+
mask = Image.fromarray(mask).convert("L") # Ensure grayscale
|
60 |
+
mask = mask.resize((512, 512))
|
61 |
+
return prepare_image(mask, (512, 512))
|
62 |
+
|
63 |
+
def inpaint_image(image, mask):
|
64 |
+
"""Performs inpainting using the LaMa model."""
|
65 |
+
outputs = lama_model.run(None, {'image': image, 'mask': mask})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
output = outputs[0][0]
|
|
|
67 |
output = output.transpose(1, 2, 0)
|
68 |
+
output = (output * 255).astype(np.uint8)
|
69 |
+
return Image.fromarray(output)
|
70 |
+
|
71 |
+
# --- Gradio Interface ---
|
72 |
+
|
73 |
+
def process_image(input_image, mask_option):
|
74 |
+
"""Main function for Gradio interface."""
|
75 |
+
imageio.imwrite("./data/data.png", input_image)
|
76 |
+
|
77 |
+
image = prepare_image(input_image)
|
78 |
+
mask = generate_mask(input_image, method=mask_option)
|
79 |
+
|
80 |
+
inpainted_image = inpaint_image(image, mask)
|
81 |
+
inpainted_image = inpainted_image.resize(Image.open("./data/data.png").size)
|
82 |
+
inpainted_image.save("./dataout/data_mask.png")
|
83 |
+
return "./dataout/data_mask.png", "./data/data_mask.png"
|
84 |
+
|
85 |
+
iface = gr.Interface(
|
86 |
+
fn=process_image,
|
87 |
+
inputs=[
|
88 |
+
gr.Image(label="Input Image", type="numpy"),
|
89 |
+
gr.Radio(choices=["automatic", "manual"],
|
90 |
+
type="value", default="manual", label="Masking Option")
|
91 |
+
],
|
92 |
+
outputs=[
|
93 |
+
gr.Image(type="file", label="Inpainted Image"),
|
94 |
+
gr.Image(type="file", label="Generated Mask")
|
95 |
+
],
|
96 |
+
title="LaMa Image Inpainting",
|
97 |
+
description="Image inpainting with LaMa and U^2-Net. Upload your image and choose automatic or manual masking.",
|
98 |
+
)
|
99 |
+
|
100 |
+
iface.launch()
|