Spaces:
Running
Running
zhang-ziang
commited on
Commit
·
b03b419
1
Parent(s):
c1fa1ed
rm_bkg
Browse files- app.py +70 -6
- requirements.txt +1 -1
app.py
CHANGED
@@ -8,6 +8,9 @@ import os
|
|
8 |
import matplotlib.pyplot as plt
|
9 |
import io
|
10 |
from PIL import Image
|
|
|
|
|
|
|
11 |
|
12 |
from huggingface_hub import hf_hub_download
|
13 |
ckpt_path = hf_hub_download(repo_id="Viglong/OriNet", filename="celarge/dino_weight.pt", repo_type="model", cache_dir='./')
|
@@ -30,6 +33,68 @@ dino.load_state_dict(torch.load(ckpt_path, map_location='cpu'))
|
|
30 |
print('weight loaded')
|
31 |
val_preprocess = AutoImageProcessor.from_pretrained(DINO_LARGE, cache_dir='./')
|
32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
def get_3angle(image):
|
35 |
|
@@ -80,10 +145,8 @@ def figure_to_img(fig):
|
|
80 |
image = Image.open(buf).copy()
|
81 |
return image
|
82 |
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
def generate_mutimodal(img):
|
87 |
angles = get_3angle(img)
|
88 |
|
89 |
fig, ax = plt.subplots(figsize=(8, 8))
|
@@ -123,9 +186,10 @@ def generate_mutimodal(img):
|
|
123 |
|
124 |
server = gr.Interface(
|
125 |
flagging_mode='never',
|
126 |
-
fn=
|
127 |
inputs=[
|
128 |
-
gr.Image(height=512, width=512, label="upload your image")
|
|
|
129 |
],
|
130 |
outputs=[
|
131 |
gr.Image(height=512, width=512, label="result image"),
|
|
|
8 |
import matplotlib.pyplot as plt
|
9 |
import io
|
10 |
from PIL import Image
|
11 |
+
import rembg
|
12 |
+
from typing import Any
|
13 |
+
|
14 |
|
15 |
from huggingface_hub import hf_hub_download
|
16 |
ckpt_path = hf_hub_download(repo_id="Viglong/OriNet", filename="celarge/dino_weight.pt", repo_type="model", cache_dir='./')
|
|
|
33 |
print('weight loaded')
|
34 |
val_preprocess = AutoImageProcessor.from_pretrained(DINO_LARGE, cache_dir='./')
|
35 |
|
36 |
+
def background_preprocess(input_image, do_remove_background):
|
37 |
+
|
38 |
+
rembg_session = rembg.new_session() if do_remove_background else None
|
39 |
+
|
40 |
+
if do_remove_background:
|
41 |
+
input_image = remove_background(input_image, rembg_session)
|
42 |
+
input_image = resize_foreground(input_image, 0.85)
|
43 |
+
|
44 |
+
return input_image
|
45 |
+
|
46 |
+
def resize_foreground(
|
47 |
+
image: Image,
|
48 |
+
ratio: float,
|
49 |
+
) -> Image:
|
50 |
+
image = np.array(image)
|
51 |
+
assert image.shape[-1] == 4
|
52 |
+
alpha = np.where(image[..., 3] > 0)
|
53 |
+
y1, y2, x1, x2 = (
|
54 |
+
alpha[0].min(),
|
55 |
+
alpha[0].max(),
|
56 |
+
alpha[1].min(),
|
57 |
+
alpha[1].max(),
|
58 |
+
)
|
59 |
+
# crop the foreground
|
60 |
+
fg = image[y1:y2, x1:x2]
|
61 |
+
# pad to square
|
62 |
+
size = max(fg.shape[0], fg.shape[1])
|
63 |
+
ph0, pw0 = (size - fg.shape[0]) // 2, (size - fg.shape[1]) // 2
|
64 |
+
ph1, pw1 = size - fg.shape[0] - ph0, size - fg.shape[1] - pw0
|
65 |
+
new_image = np.pad(
|
66 |
+
fg,
|
67 |
+
((ph0, ph1), (pw0, pw1), (0, 0)),
|
68 |
+
mode="constant",
|
69 |
+
constant_values=((0, 0), (0, 0), (0, 0)),
|
70 |
+
)
|
71 |
+
|
72 |
+
# compute padding according to the ratio
|
73 |
+
new_size = int(new_image.shape[0] / ratio)
|
74 |
+
# pad to size, double side
|
75 |
+
ph0, pw0 = (new_size - size) // 2, (new_size - size) // 2
|
76 |
+
ph1, pw1 = new_size - size - ph0, new_size - size - pw0
|
77 |
+
new_image = np.pad(
|
78 |
+
new_image,
|
79 |
+
((ph0, ph1), (pw0, pw1), (0, 0)),
|
80 |
+
mode="constant",
|
81 |
+
constant_values=((0, 0), (0, 0), (0, 0)),
|
82 |
+
)
|
83 |
+
new_image = Image.fromarray(new_image)
|
84 |
+
return new_image
|
85 |
+
|
86 |
+
def remove_background(image: Image,
|
87 |
+
rembg_session: Any = None,
|
88 |
+
force: bool = False,
|
89 |
+
**rembg_kwargs,
|
90 |
+
) -> Image:
|
91 |
+
do_remove = True
|
92 |
+
if image.mode == "RGBA" and image.getextrema()[3][0] < 255:
|
93 |
+
do_remove = False
|
94 |
+
do_remove = do_remove or force
|
95 |
+
if do_remove:
|
96 |
+
image = rembg.remove(image, session=rembg_session, **rembg_kwargs)
|
97 |
+
return image
|
98 |
|
99 |
def get_3angle(image):
|
100 |
|
|
|
145 |
image = Image.open(buf).copy()
|
146 |
return image
|
147 |
|
148 |
+
def infer_func(img, do_rm_bkg):
|
149 |
+
img = background_preprocess(img, do_rm_bkg)
|
|
|
|
|
150 |
angles = get_3angle(img)
|
151 |
|
152 |
fig, ax = plt.subplots(figsize=(8, 8))
|
|
|
186 |
|
187 |
server = gr.Interface(
|
188 |
flagging_mode='never',
|
189 |
+
fn=infer_func,
|
190 |
inputs=[
|
191 |
+
gr.Image(height=512, width=512, label="upload your image"),
|
192 |
+
gr.Checkbox(label="Remove Background", value=True)
|
193 |
],
|
194 |
outputs=[
|
195 |
gr.Image(height=512, width=512, label="result image"),
|
requirements.txt
CHANGED
@@ -5,4 +5,4 @@ pillow==10.2.0
|
|
5 |
huggingface-hub==0.26.5
|
6 |
gradio==5.9.0
|
7 |
numpy==1.26.4
|
8 |
-
|
|
|
5 |
huggingface-hub==0.26.5
|
6 |
gradio==5.9.0
|
7 |
numpy==1.26.4
|
8 |
+
rembg
|