Spaces:
Running
Running
edyoshikun
commited on
Commit
•
1f6e4a8
1
Parent(s):
6b043c2
custom CSS and adding title, description and references
Browse files
app.py
CHANGED
@@ -1,30 +1,28 @@
|
|
1 |
-
from viscy.light.engine import VSUNet
|
2 |
-
import torch
|
3 |
import gradio as gr
|
4 |
-
import
|
|
|
|
|
5 |
from numpy.typing import ArrayLike
|
|
|
6 |
from skimage import exposure
|
7 |
-
from huggingface_hub import hf_hub_download
|
8 |
|
9 |
|
10 |
class VSGradio:
|
11 |
def __init__(self, model_config, model_ckpt_path):
|
12 |
self.model_config = model_config
|
13 |
self.model_ckpt_path = model_ckpt_path
|
14 |
-
self.device = torch.device(
|
15 |
-
"cuda" if torch.cuda.is_available() else "cpu"
|
16 |
-
) # Check if GPU is available
|
17 |
self.model = None
|
18 |
self.load_model()
|
19 |
|
20 |
def load_model(self):
|
21 |
-
# Load the model checkpoint
|
22 |
self.model = VSUNet.load_from_checkpoint(
|
23 |
self.model_ckpt_path,
|
24 |
architecture="UNeXt2_2D",
|
25 |
model_config=self.model_config,
|
26 |
)
|
27 |
-
self.model.to(self.device)
|
28 |
self.model.eval()
|
29 |
|
30 |
def normalize_fov(self, input: ArrayLike):
|
@@ -34,31 +32,46 @@ class VSGradio:
|
|
34 |
return (input - mean) / std
|
35 |
|
36 |
def predict(self, inp):
|
37 |
-
#
|
38 |
-
# ensure inp is tensor has to be a (B,C,D,H,W) tensor
|
39 |
inp = self.normalize_fov(inp)
|
40 |
inp = torch.from_numpy(np.array(inp).astype(np.float32))
|
|
|
|
|
41 |
test_dict = dict(
|
42 |
index=None,
|
43 |
source=inp.unsqueeze(0).unsqueeze(0).unsqueeze(0).to(self.device),
|
44 |
)
|
|
|
|
|
45 |
with torch.inference_mode():
|
46 |
-
self.model.on_predict_start()
|
47 |
-
pred =
|
48 |
-
|
|
|
|
|
|
|
49 |
nuc_pred = pred[0, 0, 0]
|
50 |
mem_pred = pred[0, 1, 0]
|
51 |
nuc_pred = exposure.rescale_intensity(nuc_pred, out_range=(0, 1))
|
52 |
mem_pred = exposure.rescale_intensity(mem_pred, out_range=(0, 1))
|
|
|
53 |
return nuc_pred, mem_pred
|
54 |
|
55 |
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
# %%
|
57 |
if __name__ == "__main__":
|
|
|
58 |
model_ckpt_path = hf_hub_download(
|
59 |
repo_id="compmicro-czb/VSCyto2D", filename="epoch=399-step=23200.ckpt"
|
60 |
)
|
61 |
|
|
|
62 |
model_config = {
|
63 |
"in_channels": 1,
|
64 |
"out_channels": 2,
|
@@ -70,17 +83,62 @@ if __name__ == "__main__":
|
|
70 |
"pretraining": False,
|
71 |
}
|
72 |
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from viscy.light.engine import VSUNet
|
4 |
+
from huggingface_hub import hf_hub_download
|
5 |
from numpy.typing import ArrayLike
|
6 |
+
import numpy as np
|
7 |
from skimage import exposure
|
|
|
8 |
|
9 |
|
10 |
class VSGradio:
|
11 |
def __init__(self, model_config, model_ckpt_path):
|
12 |
self.model_config = model_config
|
13 |
self.model_ckpt_path = model_ckpt_path
|
14 |
+
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
|
|
15 |
self.model = None
|
16 |
self.load_model()
|
17 |
|
18 |
def load_model(self):
|
19 |
+
# Load the model checkpoint and move it to the correct device (GPU or CPU)
|
20 |
self.model = VSUNet.load_from_checkpoint(
|
21 |
self.model_ckpt_path,
|
22 |
architecture="UNeXt2_2D",
|
23 |
model_config=self.model_config,
|
24 |
)
|
25 |
+
self.model.to(self.device) # Move the model to the correct device (GPU/CPU)
|
26 |
self.model.eval()
|
27 |
|
28 |
def normalize_fov(self, input: ArrayLike):
|
|
|
32 |
return (input - mean) / std
|
33 |
|
34 |
def predict(self, inp):
|
35 |
+
# Normalize the input and convert to tensor
|
|
|
36 |
inp = self.normalize_fov(inp)
|
37 |
inp = torch.from_numpy(np.array(inp).astype(np.float32))
|
38 |
+
|
39 |
+
# Prepare the input dictionary and move input to the correct device (GPU or CPU)
|
40 |
test_dict = dict(
|
41 |
index=None,
|
42 |
source=inp.unsqueeze(0).unsqueeze(0).unsqueeze(0).to(self.device),
|
43 |
)
|
44 |
+
|
45 |
+
# Run model inference
|
46 |
with torch.inference_mode():
|
47 |
+
self.model.on_predict_start() # Necessary preprocessing for the model
|
48 |
+
pred = (
|
49 |
+
self.model.predict_step(test_dict, 0, 0).cpu().numpy()
|
50 |
+
) # Move output back to CPU for post-processing
|
51 |
+
|
52 |
+
# Post-process the model output and rescale intensity
|
53 |
nuc_pred = pred[0, 0, 0]
|
54 |
mem_pred = pred[0, 1, 0]
|
55 |
nuc_pred = exposure.rescale_intensity(nuc_pred, out_range=(0, 1))
|
56 |
mem_pred = exposure.rescale_intensity(mem_pred, out_range=(0, 1))
|
57 |
+
|
58 |
return nuc_pred, mem_pred
|
59 |
|
60 |
|
61 |
+
# Load the custom CSS from the file
|
62 |
+
def load_css(file_path):
|
63 |
+
with open(file_path, "r") as file:
|
64 |
+
return file.read()
|
65 |
+
|
66 |
+
|
67 |
# %%
|
68 |
if __name__ == "__main__":
|
69 |
+
# Download the model checkpoint from Hugging Face
|
70 |
model_ckpt_path = hf_hub_download(
|
71 |
repo_id="compmicro-czb/VSCyto2D", filename="epoch=399-step=23200.ckpt"
|
72 |
)
|
73 |
|
74 |
+
# Model configuration
|
75 |
model_config = {
|
76 |
"in_channels": 1,
|
77 |
"out_channels": 2,
|
|
|
83 |
"pretraining": False,
|
84 |
}
|
85 |
|
86 |
+
# Initialize the Gradio app using Blocks
|
87 |
+
with gr.Blocks(css=load_css("style.css")) as demo:
|
88 |
+
# Title and description
|
89 |
+
gr.HTML(
|
90 |
+
"<div class='title-block'>Image Translation (Virtual Staining) of cellular landmark organelles</div>"
|
91 |
+
)
|
92 |
+
# Improved description block with better formatting
|
93 |
+
gr.HTML(
|
94 |
+
"""
|
95 |
+
<div class='description-block'>
|
96 |
+
<p><b>Model:</b> VSCyto2D</p>
|
97 |
+
<p>
|
98 |
+
<b>Input:</b> label-free image (e.g., QPI or phase contrast) <br>
|
99 |
+
<b>Output:</b> two virtually stained channels: one for the <b>nucleus</b> and one for the <b>cell membrane</b>.
|
100 |
+
</p>
|
101 |
+
<p>
|
102 |
+
Check out our preprint:
|
103 |
+
<a href='https://www.biorxiv.org/content/10.1101/2024.05.31.596901' target='_blank'><i>Liu et al.,Robust virtual staining of landmark organelles</i></a>
|
104 |
+
</p>
|
105 |
+
</div>
|
106 |
+
"""
|
107 |
+
)
|
108 |
+
|
109 |
+
vsgradio = VSGradio(model_config, model_ckpt_path)
|
110 |
+
|
111 |
+
# Layout for input and output images
|
112 |
+
with gr.Row():
|
113 |
+
input_image = gr.Image(type="numpy", image_mode="L", label="Upload Image")
|
114 |
+
with gr.Column():
|
115 |
+
output_nucleus = gr.Image(type="numpy", label="VS Nucleus")
|
116 |
+
output_membrane = gr.Image(type="numpy", label="VS Membrane")
|
117 |
+
|
118 |
+
# Button to trigger prediction
|
119 |
+
submit_button = gr.Button("Submit")
|
120 |
+
|
121 |
+
# Define what happens when the button is clicked
|
122 |
+
submit_button.click(
|
123 |
+
vsgradio.predict,
|
124 |
+
inputs=input_image,
|
125 |
+
outputs=[output_nucleus, output_membrane],
|
126 |
+
)
|
127 |
+
|
128 |
+
# Example images and article
|
129 |
+
gr.Examples(
|
130 |
+
examples=["examples/a549.png", "examples/hek.png"], inputs=input_image
|
131 |
+
)
|
132 |
+
|
133 |
+
# Article or footer information
|
134 |
+
gr.HTML(
|
135 |
+
"""
|
136 |
+
<div class='article-block'>
|
137 |
+
<p> Model trained primarily on HEK293T, BJ5, and A549 cells. For best results, use quantitative phase images (QPI) or Zernike phase contrast.</p>
|
138 |
+
<p> For training, inference and evaluation of the model refer to the <a href='https://github.com/mehta-lab/VisCy/tree/main/examples/virtual_staining/dlmbl_exercise' target='_blank'>GitHub repository</a>.</p>
|
139 |
+
</div>
|
140 |
+
"""
|
141 |
+
)
|
142 |
+
|
143 |
+
# Launch the Gradio app
|
144 |
+
demo.launch()
|
style.css
ADDED
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
/* Default styling for light mode */
|
2 |
+
.title-block, .description-block, .article-block {
|
3 |
+
background-color: #f0f0f0; /* Light background for light mode */
|
4 |
+
border-radius: 10px;
|
5 |
+
padding: 20px;
|
6 |
+
margin-bottom: 20px;
|
7 |
+
text-align: center;
|
8 |
+
}
|
9 |
+
|
10 |
+
.title-block {
|
11 |
+
font-size: 28px;
|
12 |
+
font-weight: bold;
|
13 |
+
color: #333; /* Dark text for light mode */
|
14 |
+
}
|
15 |
+
|
16 |
+
.description-block {
|
17 |
+
font-size: 18px;
|
18 |
+
color: #444; /* Slightly lighter text for light mode */
|
19 |
+
}
|
20 |
+
|
21 |
+
.article-block {
|
22 |
+
font-size: 16px;
|
23 |
+
margin-top: 30px;
|
24 |
+
color: #555; /* Even lighter text for light mode */
|
25 |
+
}
|
26 |
+
|
27 |
+
/* Dark mode styling */
|
28 |
+
@media (prefers-color-scheme: dark) {
|
29 |
+
.title-block, .description-block, .article-block {
|
30 |
+
background-color: #2b2b2b; /* Dark background for dark mode */
|
31 |
+
color: #f0f0f0; /* Light text for dark mode */
|
32 |
+
}
|
33 |
+
|
34 |
+
.title-block {
|
35 |
+
color: #e0e0e0; /* Light text for dark mode */
|
36 |
+
}
|
37 |
+
|
38 |
+
.description-block {
|
39 |
+
color: #d0d0d0; /* Lighter text for dark mode */
|
40 |
+
}
|
41 |
+
|
42 |
+
.article-block {
|
43 |
+
color: #c0c0c0; /* Even lighter text for dark mode */
|
44 |
+
}
|
45 |
+
}
|