Spaces:
Runtime error
Runtime error
Create new file
Browse files- app_onnx.py +43 -0
app_onnx.py
ADDED
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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 |
+
def create_model_for_provider(model_path, provider="CPUExecutionProvider"):
|
8 |
+
options = ort.SessionOptions()
|
9 |
+
options.intra_op_num_threads = 1
|
10 |
+
options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
|
11 |
+
session = ort.InferenceSession(str(model_path), options, providers=[provider])
|
12 |
+
session.disable_fallback()
|
13 |
+
return session
|
14 |
+
|
15 |
+
def inference(repo_id, model_name, img):
|
16 |
+
model = hf_hub_download(repo_id=repo_id, filename=model_name)
|
17 |
+
ort_session = create_model_for_provider(model)
|
18 |
+
n_channels = ort_session.get_inputs()[0].shape[-1]
|
19 |
+
|
20 |
+
img = img[...,:n_channels]/255
|
21 |
+
ort_inputs = {ort_session.get_inputs()[0].name: img.astype(np.float32)}
|
22 |
+
|
23 |
+
ort_outs = ort_session.run(None, ort_inputs)
|
24 |
+
|
25 |
+
return ort_outs[0]*255, ort_outs[2]/0.25
|
26 |
+
|
27 |
+
title="deepflash2"
|
28 |
+
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.'
|
29 |
+
examples=[['matjesg/deepflash2_demo', 'cFOS_ensemble.onnx', 'cFOS_example.png'],
|
30 |
+
['matjesg/deepflash2_demo', 'YFP_ensemble.onnx', 'YFP_example.png']
|
31 |
+
]
|
32 |
+
|
33 |
+
gr.Interface(inference,
|
34 |
+
[gr.inputs.Textbox(placeholder='e.g., matjesg/cFOS_in_HC', label='repo_id'),
|
35 |
+
gr.inputs.Textbox(placeholder='e.g., ensemble.onnx', label='model_name'),
|
36 |
+
gr.inputs.Image(type='numpy', label='Input image')
|
37 |
+
],
|
38 |
+
[gr.outputs.Image(label='Segmentation Mask'),
|
39 |
+
gr.outputs.Image(label='Uncertainty Map')],
|
40 |
+
title=title,
|
41 |
+
description=description,
|
42 |
+
examples=examples,
|
43 |
+
).launch()
|