Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -4,7 +4,7 @@ import numpy as np
|
|
4 |
from PIL import Image
|
5 |
import os
|
6 |
|
7 |
-
# === Fix font/matplotlib warnings ===
|
8 |
os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"
|
9 |
os.environ["XDG_CACHE_HOME"] = "/tmp"
|
10 |
|
@@ -14,7 +14,8 @@ def weighted_dice_loss(y_true, y_pred):
|
|
14 |
y_true_f = tf.reshape(y_true, [-1])
|
15 |
y_pred_f = tf.reshape(y_pred, [-1])
|
16 |
intersection = tf.reduce_sum(y_true_f * y_pred_f)
|
17 |
-
return 1 - ((2. * intersection + smooth) /
|
|
|
18 |
|
19 |
def iou_metric(y_true, y_pred):
|
20 |
y_true = tf.cast(y_true > 0.5, tf.float32)
|
@@ -26,7 +27,7 @@ def iou_metric(y_true, y_pred):
|
|
26 |
def bce_loss(y_true, y_pred):
|
27 |
return tf.keras.losses.binary_crossentropy(y_true, y_pred)
|
28 |
|
29 |
-
# === Load
|
30 |
model_path = "final_model_after_third_iteration_WDL0.07_0.5155/"
|
31 |
@st.cache_resource
|
32 |
def load_model():
|
@@ -41,35 +42,51 @@ def load_model():
|
|
41 |
|
42 |
model = load_model()
|
43 |
|
44 |
-
# ===
|
45 |
-
def run_prediction(image):
|
46 |
-
image = image.convert("RGB").resize((512, 512))
|
47 |
-
x = np.expand_dims(np.array(image), axis=0)
|
48 |
-
y = model.predict(x)[0, :, :, 0]
|
49 |
-
y_norm = (y - y.min()) / (y.max() - y.min() + 1e-6)
|
50 |
-
mask = (y_norm * 255).astype(np.uint8)
|
51 |
-
return Image.fromarray(mask)
|
52 |
-
|
53 |
-
# === Streamlit UI ===
|
54 |
st.title("🕳️ Sinkhole Segmentation with EffV2-UNet")
|
55 |
|
|
|
|
|
|
|
|
|
56 |
example_dir = "examples"
|
57 |
-
example_files = sorted([
|
|
|
|
|
|
|
58 |
|
59 |
-
|
60 |
-
|
|
|
61 |
|
62 |
-
for i,
|
63 |
-
|
64 |
-
img_path = os.path.join(example_dir, filename)
|
65 |
example_img = Image.open(img_path)
|
66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
|
68 |
-
|
69 |
-
st.subheader("Original Image")
|
70 |
-
st.image(example_img, use_column_width=True)
|
71 |
|
72 |
-
|
73 |
-
|
74 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
|
|
|
4 |
from PIL import Image
|
5 |
import os
|
6 |
|
7 |
+
# === Fix font/matplotlib warnings for Hugging Face ===
|
8 |
os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"
|
9 |
os.environ["XDG_CACHE_HOME"] = "/tmp"
|
10 |
|
|
|
14 |
y_true_f = tf.reshape(y_true, [-1])
|
15 |
y_pred_f = tf.reshape(y_pred, [-1])
|
16 |
intersection = tf.reduce_sum(y_true_f * y_pred_f)
|
17 |
+
return 1 - ((2. * intersection + smooth) /
|
18 |
+
(tf.reduce_sum(y_true_f) + tf.reduce_sum(y_pred_f) + smooth))
|
19 |
|
20 |
def iou_metric(y_true, y_pred):
|
21 |
y_true = tf.cast(y_true > 0.5, tf.float32)
|
|
|
27 |
def bce_loss(y_true, y_pred):
|
28 |
return tf.keras.losses.binary_crossentropy(y_true, y_pred)
|
29 |
|
30 |
+
# === Load model ===
|
31 |
model_path = "final_model_after_third_iteration_WDL0.07_0.5155/"
|
32 |
@st.cache_resource
|
33 |
def load_model():
|
|
|
42 |
|
43 |
model = load_model()
|
44 |
|
45 |
+
# === Title ===
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
st.title("🕳️ Sinkhole Segmentation with EffV2-UNet")
|
47 |
|
48 |
+
# === File uploader ===
|
49 |
+
uploaded_image = st.file_uploader("Upload an image", type=["png", "jpg", "jpeg", "tif", "tiff"])
|
50 |
+
|
51 |
+
# === Example selector ===
|
52 |
example_dir = "examples"
|
53 |
+
example_files = sorted([
|
54 |
+
f for f in os.listdir(example_dir)
|
55 |
+
if f.lower().endswith((".jpg", ".jpeg", ".png", ".tif", ".tiff"))
|
56 |
+
])
|
57 |
|
58 |
+
if example_files:
|
59 |
+
st.subheader("🖼️ Try with an Example Image")
|
60 |
+
cols = st.columns(min(len(example_files), 4)) # up to 4 per row
|
61 |
|
62 |
+
for i, file in enumerate(example_files):
|
63 |
+
img_path = os.path.join(example_dir, file)
|
|
|
64 |
example_img = Image.open(img_path)
|
65 |
+
with cols[i % len(cols)]:
|
66 |
+
if st.button(file, key=file):
|
67 |
+
uploaded_image = img_path # simulate upload
|
68 |
+
image = example_img.convert("RGB")
|
69 |
+
st.image(image, caption=f"Example: {file}", use_column_width=True)
|
70 |
+
|
71 |
+
# === Prediction ===
|
72 |
+
if uploaded_image:
|
73 |
+
if isinstance(uploaded_image, str):
|
74 |
+
image = Image.open(uploaded_image).convert("RGB")
|
75 |
+
else:
|
76 |
+
image = Image.open(uploaded_image).convert("RGB")
|
77 |
|
78 |
+
st.image(image, caption="Input Image", use_column_width=True)
|
|
|
|
|
79 |
|
80 |
+
# Preprocess and predict
|
81 |
+
resized = image.resize((512, 512))
|
82 |
+
x = np.expand_dims(np.array(resized), axis=0)
|
83 |
+
y = model.predict(x)[0, :, :, 0]
|
84 |
+
|
85 |
+
st.text(f"Prediction min/max: {y.min():.5f} / {y.max():.5f}")
|
86 |
+
|
87 |
+
# Normalize output
|
88 |
+
y_norm = (y - y.min()) / (y.max() - y.min() + 1e-6)
|
89 |
+
mask = (y_norm * 255).astype(np.uint8)
|
90 |
+
result = Image.fromarray(mask)
|
91 |
|
92 |
+
st.image(result, caption="Predicted Segmentation", use_column_width=True)
|