Spaces:
Running
on
Zero
Running
on
Zero
ZhengPeng7
commited on
Commit
•
b59df1c
1
Parent(s):
bf97864
Add the weights option.
Browse files
app.py
CHANGED
@@ -5,6 +5,7 @@ import numpy as np
|
|
5 |
from PIL import Image
|
6 |
import torch
|
7 |
from torchvision import transforms
|
|
|
8 |
import gradio as gr
|
9 |
import spaces
|
10 |
from gradio_imageslider import ImageSlider
|
@@ -34,9 +35,9 @@ class ImagePreprocessor():
|
|
34 |
return image
|
35 |
|
36 |
|
37 |
-
|
38 |
from transformers import AutoModelForImageSegmentation
|
39 |
-
|
|
|
40 |
birefnet.to(device)
|
41 |
birefnet.eval()
|
42 |
|
@@ -44,7 +45,12 @@ birefnet.eval()
|
|
44 |
# def predict(image_1, image_2):
|
45 |
# images = [image_1, image_2]
|
46 |
@spaces.GPU
|
47 |
-
def predict(image, resolution):
|
|
|
|
|
|
|
|
|
|
|
48 |
resolution = f"{image.shape[1]}x{image.shape[0]}" if resolution == '' else resolution
|
49 |
# Image is a RGB numpy array.
|
50 |
resolution = [int(int(reso)//32*32) for reso in resolution.strip().split('x')]
|
@@ -84,7 +90,11 @@ examples[-1][1] = '512x512'
|
|
84 |
|
85 |
demo = gr.Interface(
|
86 |
fn=predict,
|
87 |
-
inputs=[
|
|
|
|
|
|
|
|
|
88 |
outputs=ImageSlider(),
|
89 |
examples=examples,
|
90 |
title='Online demo for `Bilateral Reference for High-Resolution Dichotomous Image Segmentation`',
|
|
|
5 |
from PIL import Image
|
6 |
import torch
|
7 |
from torchvision import transforms
|
8 |
+
from transformers import AutoModelForImageSegmentation
|
9 |
import gradio as gr
|
10 |
import spaces
|
11 |
from gradio_imageslider import ImageSlider
|
|
|
35 |
return image
|
36 |
|
37 |
|
|
|
38 |
from transformers import AutoModelForImageSegmentation
|
39 |
+
model_path = 'zhengpeng7/BiRefNet'
|
40 |
+
birefnet = AutoModelForImageSegmentation.from_pretrained(model_path, trust_remote_code=True)
|
41 |
birefnet.to(device)
|
42 |
birefnet.eval()
|
43 |
|
|
|
45 |
# def predict(image_1, image_2):
|
46 |
# images = [image_1, image_2]
|
47 |
@spaces.GPU
|
48 |
+
def predict(image, resolution, weights_file):
|
49 |
+
# Load BiRefNet with chosen weights
|
50 |
+
birefnet = AutoModelForImageSegmentation.from_pretrained(weights_file, trust_remote_code=True)
|
51 |
+
birefnet.to(device)
|
52 |
+
birefnet.eval()
|
53 |
+
|
54 |
resolution = f"{image.shape[1]}x{image.shape[0]}" if resolution == '' else resolution
|
55 |
# Image is a RGB numpy array.
|
56 |
resolution = [int(int(reso)//32*32) for reso in resolution.strip().split('x')]
|
|
|
90 |
|
91 |
demo = gr.Interface(
|
92 |
fn=predict,
|
93 |
+
inputs=[
|
94 |
+
'image',
|
95 |
+
gr.Textbox(lines=1, placeholder="Type the resolution (`WxH`) you want, e.g., `512x512`. Higher resolutions can be much slower for inference.", label="Resolution"),
|
96 |
+
gr.Checkbox(['zhengpeng7/BiRefNet', 'zhengpeng7/BiRefNet-portrait'], label="Models", info="Choose the weights you want.")
|
97 |
+
],
|
98 |
outputs=ImageSlider(),
|
99 |
examples=examples,
|
100 |
title='Online demo for `Bilateral Reference for High-Resolution Dichotomous Image Segmentation`',
|