Spaces:
Sleeping
Sleeping
Update src/streamlit_app.py
Browse files- src/streamlit_app.py +32 -12
src/streamlit_app.py
CHANGED
|
@@ -7,7 +7,8 @@ import tensorflow as tf
|
|
| 7 |
from tensorflow.keras.applications.resnet50 import preprocess_input
|
| 8 |
|
| 9 |
# ----------------- SETTINGS -----------------
|
| 10 |
-
|
|
|
|
| 11 |
IMG_SIZE = (224, 224)
|
| 12 |
|
| 13 |
CLASS_NAMES = [
|
|
@@ -23,10 +24,10 @@ CLASS_NAMES = [
|
|
| 23 |
def load_resnet_model():
|
| 24 |
try:
|
| 25 |
model = tf.keras.models.load_model(MODEL_PATH, compile=False)
|
|
|
|
| 26 |
except Exception as e:
|
| 27 |
st.error(f"Error loading model: {e}")
|
| 28 |
-
|
| 29 |
-
return model
|
| 30 |
|
| 31 |
|
| 32 |
def preprocess_image(image: Image.Image) -> np.ndarray:
|
|
@@ -40,16 +41,26 @@ def preprocess_image(image: Image.Image) -> np.ndarray:
|
|
| 40 |
|
| 41 |
|
| 42 |
def predict_image(model, image_array: np.ndarray):
|
|
|
|
| 43 |
preds = model.predict(image_array, verbose=0)[0]
|
| 44 |
probs = tf.nn.softmax(preds).numpy()
|
| 45 |
-
pred_idx = int(np.argmax(probs))
|
| 46 |
-
pred_class = CLASS_NAMES[pred_idx]
|
| 47 |
-
pred_conf = float(probs[pred_idx])
|
| 48 |
|
| 49 |
-
|
| 50 |
-
|
|
|
|
| 51 |
|
| 52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
|
| 54 |
|
| 55 |
# ----------------- PAGE CONFIG -----------------
|
|
@@ -95,7 +106,16 @@ if uploaded_file is not None:
|
|
| 95 |
|
| 96 |
if model is not None:
|
| 97 |
img_array = preprocess_image(image)
|
| 98 |
-
pred_class, pred_conf, probs, top3 = predict_image(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
|
| 100 |
with col2:
|
| 101 |
st.subheader("Prediction")
|
|
@@ -108,12 +128,12 @@ if uploaded_file is not None:
|
|
| 108 |
|
| 109 |
st.subheader("Class probabilities")
|
| 110 |
df = pd.DataFrame(
|
| 111 |
-
{"class": CLASS_NAMES, "probability": probs}
|
| 112 |
).set_index("class")
|
| 113 |
st.bar_chart(df)
|
| 114 |
|
| 115 |
with st.expander("Show probabilities as text"):
|
| 116 |
-
for name, p in zip(CLASS_NAMES, probs):
|
| 117 |
st.write(f"- {name}: {p:.2%}")
|
| 118 |
else:
|
| 119 |
st.info("Please upload a rice image to get a prediction.")
|
|
|
|
| 7 |
from tensorflow.keras.applications.resnet50 import preprocess_input
|
| 8 |
|
| 9 |
# ----------------- SETTINGS -----------------
|
| 10 |
+
# Falls das Modell im Root liegt, nimm: "./best_rice_resnet50.keras"
|
| 11 |
+
MODEL_PATH = "src/best_rice_resnet50.keras"
|
| 12 |
IMG_SIZE = (224, 224)
|
| 13 |
|
| 14 |
CLASS_NAMES = [
|
|
|
|
| 24 |
def load_resnet_model():
|
| 25 |
try:
|
| 26 |
model = tf.keras.models.load_model(MODEL_PATH, compile=False)
|
| 27 |
+
return model
|
| 28 |
except Exception as e:
|
| 29 |
st.error(f"Error loading model: {e}")
|
| 30 |
+
return None
|
|
|
|
| 31 |
|
| 32 |
|
| 33 |
def preprocess_image(image: Image.Image) -> np.ndarray:
|
|
|
|
| 41 |
|
| 42 |
|
| 43 |
def predict_image(model, image_array: np.ndarray):
|
| 44 |
+
"""Safe prediction with Top-3; robust gegen Längen-Mismatch."""
|
| 45 |
preds = model.predict(image_array, verbose=0)[0]
|
| 46 |
probs = tf.nn.softmax(preds).numpy()
|
|
|
|
|
|
|
|
|
|
| 47 |
|
| 48 |
+
n_model = len(probs)
|
| 49 |
+
n_labels = len(CLASS_NAMES)
|
| 50 |
+
n = min(n_model, n_labels)
|
| 51 |
|
| 52 |
+
# Hauptklasse (nur im Bereich [0, n))
|
| 53 |
+
main_idx = int(np.argmax(probs[:n]))
|
| 54 |
+
pred_class = CLASS_NAMES[main_idx]
|
| 55 |
+
pred_conf = float(probs[main_idx])
|
| 56 |
+
|
| 57 |
+
# Top-k (hier 3) nur im gültigen Bereich
|
| 58 |
+
k = min(3, n)
|
| 59 |
+
topk_idx = np.argsort(probs[:n])[-k:][::-1]
|
| 60 |
+
topk = [(CLASS_NAMES[i], float(probs[i])) for i in topk_idx]
|
| 61 |
+
|
| 62 |
+
# Nur die ersten n Wahrscheinlichkeiten zurückgeben
|
| 63 |
+
return pred_class, pred_conf, probs[:n], topk, n_model, n_labels
|
| 64 |
|
| 65 |
|
| 66 |
# ----------------- PAGE CONFIG -----------------
|
|
|
|
| 106 |
|
| 107 |
if model is not None:
|
| 108 |
img_array = preprocess_image(image)
|
| 109 |
+
pred_class, pred_conf, probs, top3, n_model, n_labels = predict_image(
|
| 110 |
+
model, img_array
|
| 111 |
+
)
|
| 112 |
+
|
| 113 |
+
# kurze Warnung, falls Modell/Labels nicht exakt matchen
|
| 114 |
+
if n_model != n_labels:
|
| 115 |
+
st.warning(
|
| 116 |
+
f"Model outputs {n_model} classes, CLASS_NAMES has {n_labels}. "
|
| 117 |
+
f"Only the first {min(n_model, n_labels)} are shown."
|
| 118 |
+
)
|
| 119 |
|
| 120 |
with col2:
|
| 121 |
st.subheader("Prediction")
|
|
|
|
| 128 |
|
| 129 |
st.subheader("Class probabilities")
|
| 130 |
df = pd.DataFrame(
|
| 131 |
+
{"class": CLASS_NAMES[: len(probs)], "probability": probs}
|
| 132 |
).set_index("class")
|
| 133 |
st.bar_chart(df)
|
| 134 |
|
| 135 |
with st.expander("Show probabilities as text"):
|
| 136 |
+
for name, p in zip(CLASS_NAMES[: len(probs)], probs):
|
| 137 |
st.write(f"- {name}: {p:.2%}")
|
| 138 |
else:
|
| 139 |
st.info("Please upload a rice image to get a prediction.")
|