Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,10 @@
|
|
1 |
import numpy as np
|
2 |
import gradio as gr
|
|
|
3 |
import onnxruntime as ort
|
4 |
from matplotlib import pyplot as plt
|
5 |
from huggingface_hub import hf_hub_download
|
6 |
|
7 |
-
model = hf_hub_download(repo_id="matjesg/cFOS_in_HC", filename="ensemble.onnx")
|
8 |
-
|
9 |
def create_model_for_provider(model_path, provider="CPUExecutionProvider"):
|
10 |
options = ort.SessionOptions()
|
11 |
options.intra_op_num_threads = 1
|
@@ -14,26 +13,29 @@ def create_model_for_provider(model_path, provider="CPUExecutionProvider"):
|
|
14 |
session.disable_fallback()
|
15 |
return session
|
16 |
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
img = img[...,:1]/255
|
22 |
|
|
|
23 |
ort_inputs = {ort_session.get_inputs()[0].name: img.astype(np.float32)}
|
24 |
|
25 |
ort_outs = ort_session.run(None, ort_inputs)
|
26 |
|
27 |
-
return ort_outs[0]*255
|
28 |
-
|
29 |
-
|
30 |
title="deepflash2"
|
31 |
-
description=
|
32 |
-
examples=[['
|
33 |
|
34 |
gr.Interface(inference,
|
35 |
-
gr.inputs.
|
36 |
-
|
|
|
|
|
|
|
|
|
37 |
title=title,
|
38 |
description=description,
|
39 |
examples=examples
|
|
|
1 |
import numpy as np
|
2 |
import gradio as gr
|
3 |
+
import imageio.v2 as imageio
|
4 |
import onnxruntime as ort
|
5 |
from matplotlib import pyplot as plt
|
6 |
from huggingface_hub import hf_hub_download
|
7 |
|
|
|
|
|
8 |
def create_model_for_provider(model_path, provider="CPUExecutionProvider"):
|
9 |
options = ort.SessionOptions()
|
10 |
options.intra_op_num_threads = 1
|
|
|
13 |
session.disable_fallback()
|
14 |
return session
|
15 |
|
16 |
+
def inference(repo_id, model_name, img):
|
17 |
+
model = hf_hub_download(repo_id=repo_id, filename=model_name)
|
18 |
+
ort_session = create_model_for_provider(model)
|
19 |
+
n_channels = ort_session.get_inputs()[0].shape[-1]
|
|
|
20 |
|
21 |
+
img = img[...,:n_channels]/255
|
22 |
ort_inputs = {ort_session.get_inputs()[0].name: img.astype(np.float32)}
|
23 |
|
24 |
ort_outs = ort_session.run(None, ort_inputs)
|
25 |
|
26 |
+
return ort_outs[0]*255, ort_outs[2]/0.25
|
27 |
+
|
|
|
28 |
title="deepflash2"
|
29 |
+
description='deepflash2 is a deep-learning pipeline for the segmentation of ambiguous microscopic images.\n deepflash2 uses deep model ensembles to achieve more accurate and reliable results. Thus, inference time will be more than a minute in this space.'
|
30 |
+
examples=[['matjesg/cFOS_in_HC', 'ensemble.onnx', '0001_cFOS.png']]
|
31 |
|
32 |
gr.Interface(inference,
|
33 |
+
[gr.inputs.Textbox(placeholder='e.g., matjesg/cFOS_in_HC', label='repo_id'),
|
34 |
+
gr.inputs.Textbox(placeholder='e.g., ensemble.onnx', label='model_name'),
|
35 |
+
gr.inputs.Image(type='numpy', label='Input image')
|
36 |
+
],
|
37 |
+
[gr.outputs.Image(label='Segmentation Mask'),
|
38 |
+
gr.outputs.Image(label='Uncertainty Map')],
|
39 |
title=title,
|
40 |
description=description,
|
41 |
examples=examples
|