Skip to content

Commit ddc22b2

Browse files
andrepedandreped
committedNov 9, 2023
Added a simple Gradio web demo
·
ShinyModel3D
1 parent 6c74cba commit ddc22b2

13 files changed

+649
-0
lines changed
 

‎.dockerfile

+8
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,8 @@
1+
venv/
2+
*.nii
3+
*.nii.gz
4+
*.pyc
5+
*.egg-info
6+
*.csv
7+
*__pycache__/
8+
*.DS_Store

‎.gitignore

+15
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,15 @@
1+
venv/
2+
*.nii
3+
*.nii.gz
4+
*.pyc
5+
*.egg-info
6+
*.csv
7+
*.ini
8+
*__pycache__/
9+
*.DS_Store
10+
*.json
11+
*.onnx
12+
*.xml
13+
*.obj
14+
*.zip
15+
*.txt

‎Dockerfile

+71
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,71 @@
1+
# read the doc: https://huggingface.co/docs/hub/spaces-sdks-docker
2+
# you will also find guides on how best to write your Dockerfile
3+
FROM python:3.8-slim
4+
5+
# set language, format and stuff
6+
ENV LANG=C.UTF-8 LC_ALL=C.UTF-8
7+
8+
WORKDIR /code
9+
10+
RUN apt-get update -y
11+
#RUN apt-get install -y python3 python3-pip
12+
RUN apt install git --fix-missing -y
13+
RUN apt install wget -y
14+
15+
# installing other libraries
16+
RUN apt-get install python3-pip -y && \
17+
apt-get -y install sudo
18+
RUN apt-get install curl -y
19+
RUN apt-get install nano -y
20+
RUN apt-get update && apt-get install -y git
21+
RUN apt-get install libblas-dev -y && apt-get install liblapack-dev -y
22+
RUN apt-get install gfortran -y
23+
RUN apt-get install libpng-dev -y
24+
RUN apt-get install python3-dev -y
25+
26+
WORKDIR /code
27+
28+
# install dependencies
29+
COPY ./demo/requirements.txt /code/demo/requirements.txt
30+
RUN pip install --no-cache-dir --upgrade -r /code/demo/requirements.txt
31+
32+
# resolve issue with tf==2.4 and gradio dependency collision issue
33+
RUN pip install --force-reinstall typing_extensions==4.7.1
34+
35+
# lower pydantic version to work with typing_extensions deprecation
36+
RUN pip install --force-reinstall "pydantic<2.0.0"
37+
38+
# Install wget
39+
RUN apt install wget -y && \
40+
apt install unzip
41+
42+
# Set up a new user named "user" with user ID 1000
43+
RUN useradd -m -u 1000 user
44+
45+
# Switch to the "user" user
46+
USER user
47+
48+
# Set home to the user's home directory
49+
ENV HOME=/home/user \
50+
PATH=/home/user/.local/bin:$PATH
51+
52+
# Set the working directory to the user's home directory
53+
WORKDIR $HOME/app
54+
55+
# Copy the current directory contents into the container at $HOME/app setting the owner to the user
56+
COPY --chown=user . $HOME/app
57+
58+
# Download pretrained models
59+
RUN wget "https://github.com/raidionics/Raidionics-models/releases/download/1.2.0/Raidionics-CT_LymphNodes-ONNX-v12.zip" && \
60+
unzip "Raidionics-CT_LymphNodes-ONNX-v12.zip" && mkdir -p resources/models/ && mv CT_LymphNodes/ resources/models/CT_LymphNodes/
61+
RUN wget "https://github.com/raidionics/Raidionics-models/releases/download/1.2.0/Raidionics-CT_Lungs-ONNX-v12.zip" && \
62+
unzip "Raidionics-CT_Lungs-ONNX-v12.zip" && mv CT_Lungs/ resources/models/CT_Lungs/
63+
64+
RUN rm -r *.zip
65+
66+
# Download test sample
67+
# @TODO: I have resampled the volume to 1mm isotropic for faster computation
68+
RUN wget "https://github.com/andreped/neukit/releases/download/test-data/test_thorax_CT.nii.gz"
69+
70+
# CMD ["/bin/bash"]
71+
CMD ["python3", "demo/app.py"]

‎demo/README.md

+27
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,27 @@
1+
# Hugging Face demo - through docker SDK
2+
3+
Deploying simple models in a gradio-based web interface in Hugging Face spaces is easy.
4+
For any other custom pipeline, with various dependencies and challenging behaviour, it
5+
might be necessary to use Docker containers instead.
6+
7+
For every new push to the main branch, continuous deployment to the Hugging Face
8+
`LyNoS` space is performed through a GitHub Actions workflow.
9+
10+
When the space is updated, the Docker image is rebuilt/updated (caching if possible).
11+
Then when finished, the end users can test the app as they please.
12+
13+
Right now, the functionality of the app is extremely limited, only offering a widget
14+
for uploading a NIfTI file (`.nii` or `.nii.gz`) and visualizing the produced surface
15+
of the predicted lung tumor volume when finished processing.
16+
17+
Analysis process can be monitored from the `Logs` tab next to the `Running` button
18+
in the Hugging Face `LyNoS` space.
19+
20+
It is also possible to build the app as a docker image and deploy it. To do so follow these steps:
21+
22+
```
23+
docker build -t LyNoS:latest ..
24+
docker run -it -p 7860:7860 LyNoS:latest
25+
```
26+
27+
Then open `http://localhost:7860` in your favourite internet browser to view the demo.

‎demo/app.py

+43
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,43 @@
1+
import os
2+
from argparse import ArgumentParser
3+
4+
from src.gui import WebUI
5+
6+
7+
def main():
8+
parser = ArgumentParser()
9+
parser.add_argument(
10+
"--cwd",
11+
type=str,
12+
default="/home/user/app/",
13+
help="Set current working directory (path to app.py).",
14+
)
15+
parser.add_argument(
16+
"--share",
17+
type=int,
18+
default=0,
19+
help="Whether to enable the app to be accessible online"
20+
"-> setups a public link which requires internet access.",
21+
)
22+
args = parser.parse_args()
23+
24+
print("Current working directory:", args.cwd)
25+
26+
if not os.path.exists(args.cwd):
27+
raise ValueError("Chosen 'cwd' is not a valid path!")
28+
if args.share not in [0, 1]:
29+
raise ValueError(
30+
"The 'share' argument can only be set to 0 or 1, but was:",
31+
args.share,
32+
)
33+
34+
print("Current cwd:", args.cwd)
35+
36+
# initialize and run app
37+
print("Launching demo...")
38+
app = WebUI(cwd=args.cwd, share=args.share)
39+
app.run()
40+
41+
42+
if __name__ == "__main__":
43+
main()

‎demo/src/__init__.py

Whitespace-only changes.

‎demo/src/convert.py

+35
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,35 @@
1+
import nibabel as nib
2+
from nibabel.processing import resample_to_output
3+
from skimage.measure import marching_cubes
4+
5+
6+
def nifti_to_obj(path, output="prediction.obj"):
7+
# load NIFTI into numpy array
8+
image = nib.load(path)
9+
resampled = resample_to_output(image, [1, 1, 1], order=1)
10+
data = resampled.get_fdata().astype("uint8")
11+
12+
# Create a material with a red diffuse color (RGB value)
13+
red_material = "newmtl RedMaterial\nKd 1 0 0" # Red diffuse color (RGB)
14+
15+
# extract surface
16+
verts, faces, normals, values = marching_cubes(data, 0)
17+
faces += 1
18+
19+
with open(output, "w") as thefile:
20+
# Write the material definition to the OBJ file
21+
thefile.write(red_material + "\n")
22+
23+
for item in verts:
24+
# thefile.write('usemtl RedMaterial\n')
25+
thefile.write("v {0} {1} {2}\n".format(item[0], item[1], item[2]))
26+
27+
for item in normals:
28+
thefile.write("vn {0} {1} {2}\n".format(item[0], item[1], item[2]))
29+
30+
for item in faces:
31+
thefile.write(
32+
"f {0}//{0} {1}//{1} {2}//{2}\n".format(
33+
item[0], item[1], item[2]
34+
)
35+
)

‎demo/src/css_style.py

+27
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,27 @@
1+
css = """
2+
#model-3d {
3+
height: 512px;
4+
}
5+
#model-2d {
6+
height: 512px;
7+
margin: auto;
8+
}
9+
#upload {
10+
height: 110px;
11+
}
12+
#run-button {
13+
height: 110px;
14+
width: 150px;
15+
}
16+
#toggle-button {
17+
height: 47px;
18+
width: 150px;
19+
}
20+
#logs-button {
21+
height: 47px;
22+
width: 150px;
23+
}
24+
#logs {
25+
height: auto
26+
}
27+
"""

‎demo/src/gui.py

+231
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,231 @@
1+
import os
2+
3+
import gradio as gr
4+
5+
from .convert import nifti_to_obj
6+
from .css_style import css
7+
from .inference import run_model
8+
from .logger import flush_logs
9+
from .logger import read_logs
10+
from .logger import setup_logger
11+
from .utils import load_ct_to_numpy
12+
from .utils import load_pred_volume_to_numpy
13+
14+
# setup logging
15+
LOGGER = setup_logger()
16+
17+
18+
class WebUI:
19+
def __init__(
20+
self,
21+
model_name: str = None,
22+
cwd: str = "/home/user/app/",
23+
share: int = 1,
24+
):
25+
# global states
26+
self.images = []
27+
self.pred_images = []
28+
29+
# @TODO: This should be dynamically set based on chosen volume size
30+
self.nb_slider_items = 820
31+
32+
self.model_name = model_name
33+
self.cwd = cwd
34+
self.share = share
35+
36+
self.class_name = "Lymph Nodes" # default
37+
self.class_names = {
38+
"Lymph Nodes": "CT_LymphNodes",
39+
}
40+
41+
self.result_names = {
42+
"Lymph Nodes": "LymphNodes",
43+
}
44+
45+
# define widgets not to be rendered immediantly, but later on
46+
self.slider = gr.Slider(
47+
minimum=1,
48+
maximum=self.nb_slider_items,
49+
value=1,
50+
step=1,
51+
label="Which 2D slice to show",
52+
)
53+
self.volume_renderer = gr.Model3D(
54+
clear_color=[0.0, 0.0, 0.0, 0.0],
55+
label="3D Model",
56+
show_label=True,
57+
visible=True,
58+
elem_id="model-3d",
59+
camera_position=[90, 180, 768],
60+
).style(height=512)
61+
62+
def set_class_name(self, value):
63+
LOGGER.info(f"Changed task to: {value}")
64+
self.class_name = value
65+
66+
def combine_ct_and_seg(self, img, pred):
67+
return (img, [(pred, self.class_name)])
68+
69+
def upload_file(self, file):
70+
out = file.name
71+
LOGGER.info(f"File uploaded: {out}")
72+
return out
73+
74+
def process(self, mesh_file_name):
75+
path = mesh_file_name.name
76+
run_model(
77+
path,
78+
model_path=os.path.join(self.cwd, "resources/models/"),
79+
task=self.class_names[self.class_name],
80+
name=self.result_names[self.class_name],
81+
)
82+
LOGGER.info("Converting prediction NIfTI to OBJ...")
83+
nifti_to_obj("prediction.nii.gz")
84+
85+
LOGGER.info("Loading CT to numpy...")
86+
self.images = load_ct_to_numpy(path)
87+
88+
LOGGER.info("Loading prediction volume to numpy..")
89+
self.pred_images = load_pred_volume_to_numpy("./prediction.nii.gz")
90+
91+
return "./prediction.obj"
92+
93+
def get_img_pred_pair(self, k):
94+
k = int(k)
95+
out = gr.AnnotatedImage(
96+
self.combine_ct_and_seg(self.images[k], self.pred_images[k]),
97+
visible=True,
98+
elem_id="model-2d",
99+
).style(
100+
color_map={self.class_name: "#ffae00"},
101+
height=512,
102+
width=512,
103+
)
104+
return out
105+
106+
def toggle_sidebar(self, state):
107+
state = not state
108+
return gr.update(visible=state), state
109+
110+
def run(self):
111+
with gr.Blocks(css=css) as demo:
112+
with gr.Row():
113+
with gr.Column(visible=True, scale=0.2) as sidebar_left:
114+
logs = gr.Textbox(
115+
placeholder="\n" * 16,
116+
label="Logs",
117+
info="Verbose from inference will be displayed below.",
118+
lines=38,
119+
max_lines=38,
120+
autoscroll=True,
121+
elem_id="logs",
122+
show_copy_button=True,
123+
scroll_to_output=False,
124+
container=True,
125+
line_breaks=True,
126+
)
127+
demo.load(read_logs, None, logs, every=1)
128+
129+
with gr.Column():
130+
with gr.Row():
131+
with gr.Column(scale=0.2, min_width=150):
132+
sidebar_state = gr.State(True)
133+
134+
btn_toggle_sidebar = gr.Button(
135+
"Toggle Sidebar",
136+
elem_id="toggle-button",
137+
)
138+
btn_toggle_sidebar.click(
139+
self.toggle_sidebar,
140+
[sidebar_state],
141+
[sidebar_left, sidebar_state],
142+
)
143+
144+
btn_clear_logs = gr.Button(
145+
"Clear logs", elem_id="logs-button"
146+
)
147+
btn_clear_logs.click(flush_logs, [], [])
148+
149+
file_output = gr.File(
150+
file_count="single", elem_id="upload"
151+
)
152+
file_output.upload(
153+
self.upload_file, file_output, file_output
154+
)
155+
156+
model_selector = gr.Dropdown(
157+
list(self.class_names.keys()),
158+
label="Task",
159+
info="Which structure to segment.",
160+
multiselect=False,
161+
size="sm",
162+
)
163+
model_selector.input(
164+
fn=lambda x: self.set_class_name(x),
165+
inputs=model_selector,
166+
outputs=None,
167+
)
168+
169+
with gr.Column(scale=0.2, min_width=150):
170+
run_btn = gr.Button(
171+
"Run analysis",
172+
variant="primary",
173+
elem_id="run-button",
174+
).style(
175+
full_width=False,
176+
size="lg",
177+
)
178+
run_btn.click(
179+
fn=lambda x: self.process(x),
180+
inputs=file_output,
181+
outputs=self.volume_renderer,
182+
)
183+
184+
with gr.Row():
185+
gr.Examples(
186+
examples=[
187+
os.path.join(self.cwd, "test_thorax_CT.nii.gz"),
188+
],
189+
inputs=file_output,
190+
outputs=file_output,
191+
fn=self.upload_file,
192+
cache_examples=True,
193+
)
194+
195+
gr.Markdown(
196+
"""
197+
**NOTE:** Inference might take several minutes (Lymph nodes: ~8 minutes), see logs to the left. \\
198+
The segmentation will be available in the 2D and 3D viewers below when finished.
199+
"""
200+
)
201+
202+
with gr.Row():
203+
with gr.Box():
204+
with gr.Column():
205+
# create dummy image to be replaced by loaded images
206+
t = gr.AnnotatedImage(
207+
visible=True, elem_id="model-2d"
208+
).style(
209+
color_map={self.class_name: "#ffae00"},
210+
height=512,
211+
width=512,
212+
)
213+
214+
self.slider.input(
215+
self.get_img_pred_pair,
216+
self.slider,
217+
t,
218+
)
219+
220+
self.slider.render()
221+
222+
with gr.Box():
223+
self.volume_renderer.render()
224+
225+
# sharing app publicly -> share=True:
226+
# https://gradio.app/sharing-your-app/
227+
# inference times > 60 seconds -> need queue():
228+
# https://github.com/tloen/alpaca-lora/issues/60#issuecomment-1510006062
229+
demo.queue().launch(
230+
server_name="0.0.0.0", server_port=7860, share=self.share
231+
)

‎demo/src/inference.py

+101
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,101 @@
1+
import configparser
2+
import logging
3+
import os
4+
import shutil
5+
import traceback
6+
7+
8+
def run_model(
9+
input_path: str,
10+
model_path: str,
11+
verbose: str = "info",
12+
task: str = "CT_LymphNodes",
13+
name: str = "Lymph nodes",
14+
):
15+
if verbose == "debug":
16+
logging.getLogger().setLevel(logging.DEBUG)
17+
elif verbose == "info":
18+
logging.getLogger().setLevel(logging.INFO)
19+
elif verbose == "error":
20+
logging.getLogger().setLevel(logging.ERROR)
21+
else:
22+
raise ValueError("Unsupported verbose value provided:", verbose)
23+
24+
# delete patient/result folder if they exist
25+
if os.path.exists("./patient/"):
26+
shutil.rmtree("./patient/")
27+
if os.path.exists("./result/"):
28+
shutil.rmtree("./result/")
29+
30+
patient_directory = ""
31+
output_path = ""
32+
try:
33+
# setup temporary patient directory
34+
filename = input_path.split("/")[-1]
35+
splits = filename.split(".")
36+
extension = ".".join(splits[1:])
37+
patient_directory = "./patient/"
38+
os.makedirs(patient_directory + "T0/", exist_ok=True)
39+
shutil.copy(
40+
input_path,
41+
patient_directory + "T0/" + splits[0] + "-t1gd." + extension,
42+
)
43+
44+
# define output directory to save results
45+
output_path = "./result/prediction-" + splits[0] + "/"
46+
os.makedirs(output_path, exist_ok=True)
47+
48+
# Setting up the configuration file
49+
rads_config = configparser.ConfigParser()
50+
rads_config.add_section("Default")
51+
rads_config.set("Default", "task", "mediastinum_diagnosis")
52+
rads_config.set("Default", "caller", "")
53+
rads_config.add_section("System")
54+
rads_config.set("System", "gpu_id", "-1")
55+
rads_config.set("System", "input_folder", patient_directory)
56+
rads_config.set("System", "output_folder", output_path)
57+
rads_config.set("System", "model_folder", model_path)
58+
rads_config.set(
59+
"System",
60+
"pipeline_filename",
61+
os.path.join(model_path, task, "pipeline.json"),
62+
)
63+
rads_config.add_section("Runtime")
64+
rads_config.set(
65+
"Runtime", "reconstruction_method", "thresholding"
66+
) # thresholding, probabilities
67+
rads_config.set("Runtime", "reconstruction_order", "resample_first")
68+
rads_config.set("Runtime", "use_preprocessed_data", "False")
69+
70+
with open("rads_config.ini", "w") as f:
71+
rads_config.write(f)
72+
73+
# finally, run inference
74+
from raidionicsrads.compute import run_rads
75+
76+
run_rads(config_filename="rads_config.ini")
77+
78+
# rename and move final result
79+
os.rename(
80+
"./result/prediction-"
81+
+ splits[0]
82+
+ "/T0/"
83+
+ splits[0]
84+
+ "-t1gd_annotation-"
85+
+ name
86+
+ ".nii.gz",
87+
"./prediction.nii.gz",
88+
)
89+
# Clean-up
90+
if os.path.exists(patient_directory):
91+
shutil.rmtree(patient_directory)
92+
if os.path.exists(output_path):
93+
shutil.rmtree(output_path)
94+
95+
except Exception:
96+
print(traceback.format_exc())
97+
# Clean-up
98+
if os.path.exists(patient_directory):
99+
shutil.rmtree(patient_directory)
100+
if os.path.exists(output_path):
101+
shutil.rmtree(output_path)

‎demo/src/logger.py

+37
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,37 @@
1+
import logging
2+
import sys
3+
4+
5+
def get_logger():
6+
return logging.getLogger(__name__)
7+
8+
9+
def setup_logger():
10+
# clear log
11+
file_to_delete = open("log.txt", "w")
12+
file_to_delete.close()
13+
14+
file_handler = logging.FileHandler(filename="log.txt")
15+
stdout_handler = logging.StreamHandler(stream=sys.stdout)
16+
handlers = [file_handler, stdout_handler]
17+
18+
logging.basicConfig(
19+
level=logging.INFO,
20+
format="[%(asctime)s] {%(filename)s:%(lineno)d} %(levelname)s - %(message)s",
21+
handlers=handlers,
22+
)
23+
24+
return get_logger()
25+
26+
27+
def read_logs():
28+
sys.stdout.flush()
29+
with open("log.txt", "r") as f:
30+
return f.read()
31+
32+
33+
def flush_logs():
34+
sys.stdout.flush()
35+
# clear log
36+
file_to_delete = open("log.txt", "w")
37+
file_to_delete.close()

‎demo/src/utils.py

+40
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,40 @@
1+
import nibabel as nib
2+
import numpy as np
3+
4+
5+
def load_ct_to_numpy(data_path):
6+
if not isinstance(data_path, str):
7+
data_path = data_path.name
8+
9+
image = nib.load(data_path)
10+
data = image.get_fdata()
11+
12+
data = np.rot90(data, k=1, axes=(0, 1))
13+
data = np.flip(data, axis=0)
14+
15+
data[data < -1024] = 1024
16+
data[data > 1024] = 1024
17+
18+
data = data - np.amin(data)
19+
data = data / np.amax(data) * 255
20+
data = data.astype("uint8")
21+
22+
print(data.shape)
23+
return [data[..., i] for i in range(data.shape[-1])]
24+
25+
26+
def load_pred_volume_to_numpy(data_path):
27+
if not isinstance(data_path, str):
28+
data_path = data_path.name
29+
30+
image = nib.load(data_path)
31+
data = image.get_fdata()
32+
33+
data = np.rot90(data, k=1, axes=(0, 1))
34+
data = np.flip(data, axis=0)
35+
36+
data[data > 0] = 1
37+
data = data.astype("uint8")
38+
39+
print(data.shape)
40+
return [data[..., i] for i in range(data.shape[-1])]

‎setup.cfg

+14
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,14 @@
1+
[metadata]
2+
description-file = README.md
3+
4+
[isort]
5+
force_single_line=True
6+
known_first_party=lynos
7+
line_length=160
8+
profile=black
9+
10+
[flake8]
11+
# imported but unused in __init__.py, that's ok.
12+
per-file-ignores=*__init__.py:F401
13+
ignore=E203,W503,W605,F632,E266,E731,E712,E741
14+
max-line-length=160

0 commit comments

Comments
 (0)
Please sign in to comment.