Spaces:
Runtime error
Runtime error
add: plot and bar
Browse files- app.py +32 -18
- utilities/load_model.py +19 -0
app.py
CHANGED
@@ -1,27 +1,34 @@
|
|
1 |
# import the necessary packages
|
2 |
from utilities import config
|
|
|
3 |
from tensorflow.keras import layers
|
4 |
-
from tensorflow import keras
|
5 |
import tensorflow as tf
|
6 |
import matplotlib.pyplot as plt
|
7 |
import math
|
8 |
import gradio as gr
|
9 |
|
10 |
# load the models from disk
|
11 |
-
conv_stem =
|
12 |
-
config.IMAGENETTE_STEM_PATH,
|
13 |
-
|
14 |
-
|
15 |
-
conv_trunk = keras.models.load_model(
|
16 |
-
config.IMAGENETTE_TRUNK_PATH,
|
17 |
-
compile=False
|
18 |
-
)
|
19 |
-
conv_attn = keras.models.load_model(
|
20 |
-
config.IMAGENETTE_ATTN_PATH,
|
21 |
-
compile=False
|
22 |
)
|
23 |
|
24 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
# resize the image to a 224, 224 dim
|
26 |
image = tf.image.convert_image_dtype(image, tf.float32)
|
27 |
image = tf.image.resize(image, (224, 224))
|
@@ -32,7 +39,7 @@ def plot_attention(image):
|
|
32 |
# pass through the trunk
|
33 |
test_x = conv_trunk(test_x)
|
34 |
# pass through the attention pooling block
|
35 |
-
|
36 |
test_viz_weights = test_viz_weights[tf.newaxis, ...]
|
37 |
|
38 |
# reshape the vizualization weights
|
@@ -52,9 +59,16 @@ def plot_attention(image):
|
|
52 |
extent=img.get_extent()
|
53 |
)
|
54 |
plt.axis("off")
|
55 |
-
|
|
|
|
|
|
|
56 |
|
57 |
iface = gr.Interface(
|
58 |
-
fn=
|
59 |
-
inputs=
|
60 |
-
outputs=
|
|
|
|
|
|
|
|
|
|
1 |
# import the necessary packages
|
2 |
from utilities import config
|
3 |
+
from utilities import load_model
|
4 |
from tensorflow.keras import layers
|
|
|
5 |
import tensorflow as tf
|
6 |
import matplotlib.pyplot as plt
|
7 |
import math
|
8 |
import gradio as gr
|
9 |
|
10 |
# load the models from disk
|
11 |
+
(conv_stem, conv_trunk, conv_attn) = load_model.loader(
|
12 |
+
stem=config.IMAGENETTE_STEM_PATH,
|
13 |
+
trunk=config.IMAGENETTE_TRUNK_PATH,
|
14 |
+
attn=config.IMAGENETTE_ATTN_PATH,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
)
|
16 |
|
17 |
+
# load labels
|
18 |
+
labels = [
|
19 |
+
'tench',
|
20 |
+
'english springer',
|
21 |
+
'cassette player',
|
22 |
+
'chain saw',
|
23 |
+
'church',
|
24 |
+
'french horn',
|
25 |
+
'garbage truck',
|
26 |
+
'gas pump',
|
27 |
+
'golf ball',
|
28 |
+
'parachute'
|
29 |
+
]
|
30 |
+
|
31 |
+
def get_results(image):
|
32 |
# resize the image to a 224, 224 dim
|
33 |
image = tf.image.convert_image_dtype(image, tf.float32)
|
34 |
image = tf.image.resize(image, (224, 224))
|
|
|
39 |
# pass through the trunk
|
40 |
test_x = conv_trunk(test_x)
|
41 |
# pass through the attention pooling block
|
42 |
+
logits, test_viz_weights = conv_attn(test_x)
|
43 |
test_viz_weights = test_viz_weights[tf.newaxis, ...]
|
44 |
|
45 |
# reshape the vizualization weights
|
|
|
59 |
extent=img.get_extent()
|
60 |
)
|
61 |
plt.axis("off")
|
62 |
+
|
63 |
+
prediction = tf.nn.softmax(logits, axis=-1)
|
64 |
+
|
65 |
+
return plt, {labels[i]: float(prediction[i]) for i in range(10)}
|
66 |
|
67 |
iface = gr.Interface(
|
68 |
+
fn=get_results,
|
69 |
+
inputs=gr.inputs.Image(label="Input Image"),
|
70 |
+
outputs=[
|
71 |
+
gr.outputs.Image(label="Attention Map"),
|
72 |
+
gr.outputs.Label(num_top_classes=10)
|
73 |
+
]
|
74 |
+
).launch()
|
utilities/load_model.py
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# import the necessary packages
|
2 |
+
from tensorflow import keras
|
3 |
+
|
4 |
+
def loader(stem, trunk, attn):
|
5 |
+
# load the models from disk
|
6 |
+
conv_stem = keras.models.load_model(
|
7 |
+
stem,
|
8 |
+
compile=False
|
9 |
+
)
|
10 |
+
conv_trunk = keras.models.load_model(
|
11 |
+
trunk,
|
12 |
+
compile=False
|
13 |
+
)
|
14 |
+
conv_attn = keras.models.load_model(
|
15 |
+
attn,
|
16 |
+
compile=False
|
17 |
+
)
|
18 |
+
|
19 |
+
return (conv_stem, conv_trunk, conv_attn)
|