Spaces:
Runtime error
Runtime error
Chirag1994
commited on
Commit
•
422fb0d
1
Parent(s):
845deaa
demo.launch() fix
Browse files
app.py
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
|
2 |
import os
|
3 |
import torch
|
4 |
import numpy as np
|
@@ -6,10 +6,11 @@ import gradio as gr
|
|
6 |
from model import Model
|
7 |
import albumentations as A
|
8 |
|
9 |
-
|
10 |
-
efficientnet_b5_model =
|
11 |
-
efficientnet_b5_model = torch.nn.DataParallel(
|
12 |
-
|
|
|
13 |
efficientnet_b5_model.load_state_dict(
|
14 |
torch.load(
|
15 |
f='efficientnet_b5_checkpoint_fold_0.pt',
|
@@ -17,7 +18,9 @@ efficientnet_b5_model.load_state_dict(
|
|
17 |
)
|
18 |
)
|
19 |
|
20 |
-
|
|
|
|
|
21 |
def predict_on_single_image(img):
|
22 |
"""
|
23 |
Function takes an image, transforms for
|
@@ -28,13 +31,13 @@ def predict_on_single_image(img):
|
|
28 |
having melanoma.
|
29 |
"""
|
30 |
img = np.array(img)
|
31 |
-
transforms = A.Compose([A.Resize(512,512),
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
img = transforms(image=img)['image']
|
39 |
image = np.transpose(img, (2, 0, 1)).astype(np.float32)
|
40 |
image = torch.tensor(image, dtype=torch.float).unsqueeze(dim=0)
|
@@ -43,27 +46,28 @@ def predict_on_single_image(img):
|
|
43 |
probs = torch.sigmoid(efficientnet_b5_model(image))
|
44 |
prob_of_melanoma = probs[0].item()
|
45 |
prob_of_not_having_melanoma = 1 - prob_of_melanoma
|
46 |
-
pred_label = {"Probability of Having Melanoma": prob_of_melanoma,
|
47 |
-
"Probability of Not having Melanoma": prob_of_not_having_melanoma}
|
48 |
return pred_label
|
49 |
-
|
50 |
-
## Gradio App
|
51 |
-
import gradio as gr
|
52 |
|
53 |
-
|
|
|
|
|
|
|
54 |
melanoma_app_examples_path = "examples"
|
55 |
|
56 |
-
|
57 |
title = "Melanoma Cancer Detection App"
|
58 |
description = 'An efficientnet-b5 model that predicts the probability of a patient having melanoma skin cancer or not.'
|
59 |
-
example_list = [["examples/" + example]
|
|
|
60 |
|
61 |
-
|
62 |
demo = gr.Interface(fn=predict_on_single_image,
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
|
68 |
-
|
69 |
-
demo.launch(
|
|
|
1 |
+
# Importing Libraries
|
2 |
import os
|
3 |
import torch
|
4 |
import numpy as np
|
|
|
6 |
from model import Model
|
7 |
import albumentations as A
|
8 |
|
9 |
+
# Creating a model instance
|
10 |
+
efficientnet_b5_model = Model()
|
11 |
+
efficientnet_b5_model = torch.nn.DataParallel(
|
12 |
+
efficientnet_b5_model) # Must wrap our model in nn.DataParallel()
|
13 |
+
# if used multi-gpu's to train the model otherwise we would get state_dict keys mismatch error.
|
14 |
efficientnet_b5_model.load_state_dict(
|
15 |
torch.load(
|
16 |
f='efficientnet_b5_checkpoint_fold_0.pt',
|
|
|
18 |
)
|
19 |
)
|
20 |
|
21 |
+
# Predict on a single image
|
22 |
+
|
23 |
+
|
24 |
def predict_on_single_image(img):
|
25 |
"""
|
26 |
Function takes an image, transforms for
|
|
|
31 |
having melanoma.
|
32 |
"""
|
33 |
img = np.array(img)
|
34 |
+
transforms = A.Compose([A.Resize(512, 512),
|
35 |
+
A.Normalize(mean=(0.485, 0.456, 0.406),
|
36 |
+
std=(0.229, 0.224, 0.225),
|
37 |
+
max_pixel_value=255.0,
|
38 |
+
always_apply=True
|
39 |
+
)]
|
40 |
+
)
|
41 |
img = transforms(image=img)['image']
|
42 |
image = np.transpose(img, (2, 0, 1)).astype(np.float32)
|
43 |
image = torch.tensor(image, dtype=torch.float).unsqueeze(dim=0)
|
|
|
46 |
probs = torch.sigmoid(efficientnet_b5_model(image))
|
47 |
prob_of_melanoma = probs[0].item()
|
48 |
prob_of_not_having_melanoma = 1 - prob_of_melanoma
|
49 |
+
pred_label = {"Probability of Having Melanoma": prob_of_melanoma,
|
50 |
+
"Probability of Not having Melanoma": prob_of_not_having_melanoma}
|
51 |
return pred_label
|
|
|
|
|
|
|
52 |
|
53 |
+
|
54 |
+
# Gradio App
|
55 |
+
|
56 |
+
# Examples directory path
|
57 |
melanoma_app_examples_path = "examples"
|
58 |
|
59 |
+
# Creating the title and description strings
|
60 |
title = "Melanoma Cancer Detection App"
|
61 |
description = 'An efficientnet-b5 model that predicts the probability of a patient having melanoma skin cancer or not.'
|
62 |
+
example_list = [["examples/" + example]
|
63 |
+
for example in os.listdir(melanoma_app_examples_path)]
|
64 |
|
65 |
+
# Create the Gradio demo
|
66 |
demo = gr.Interface(fn=predict_on_single_image,
|
67 |
+
inputs=gr.Image(type='pil'),
|
68 |
+
outputs=[gr.Label(label='Probabilities')],
|
69 |
+
examples=example_list, title=title,
|
70 |
+
description=description)
|
71 |
|
72 |
+
# Launch the demo!
|
73 |
+
demo.launch()
|