EnYa32 commited on
Commit
f68a63a
Β·
verified Β·
1 Parent(s): d51c4b8

Update src/streamlit_app.py

Browse files
Files changed (1) hide show
  1. src/streamlit_app.py +61 -29
src/streamlit_app.py CHANGED
@@ -1,17 +1,23 @@
1
- # --- Fish Classifier (CNN from scratch) ---
2
  import streamlit as st
3
  import numpy as np
 
4
  from PIL import Image
5
  import tensorflow as tf
6
  from tensorflow.keras.models import load_model
7
 
 
 
 
8
  st.set_page_config(
9
- page_title="Fish Classifier (CNN – No Transfer Learning)",
10
  layout="centered"
11
  )
12
 
13
- # ------------------ SETTINGS ------------------
14
- MODEL_PATH = "src/best_fish_cnn.keras"
 
 
15
  IMG_SIZE = (224, 224)
16
 
17
  CLASS_NAMES = [
@@ -26,73 +32,99 @@ CLASS_NAMES = [
26
  "Trout"
27
  ]
28
 
29
- # ------------------ CACHED MODEL LOADING ------------------
 
 
30
  @st.cache_resource
31
  def load_model_cached():
32
- return load_model(MODEL_PATH, compile=False)
 
 
33
 
34
  model = load_model_cached()
35
 
36
- # ------------------ HELPER FUNCTIONS ------------------
 
 
37
  def preprocess_image(image: Image.Image) -> np.ndarray:
38
  """Resize and normalize image for the CNN model."""
39
  image = image.convert("RGB")
40
  image = image.resize(IMG_SIZE)
41
- img_array = np.array(image).astype("float32") / 255.0 # weil du mit rescale=1./255 trainiert hast
42
- img_array = np.expand_dims(img_array, axis=0) # (1, 224, 224, 3)
43
  return img_array
44
 
45
  def predict_image(image: Image.Image):
46
- """Run model prediction and return class index and probabilities."""
47
  img_batch = preprocess_image(image)
48
  preds = model.predict(img_batch)
49
- probs = preds[0]
50
  pred_index = int(np.argmax(probs))
51
  pred_class = CLASS_NAMES[pred_index]
52
  return pred_class, probs
53
 
54
- # ------------------ STREAMLIT UI ------------------
55
- st.set_page_config(
56
- page_title="Fish Classifier (CNN – From Scratch)",
57
- layout="centered"
58
- )
59
-
60
  st.title("🐟 Fish Species Classifier (CNN From Scratch)")
 
61
  st.write(
62
  "Upload a fish image and the model will predict one of **9 fish species**. "
63
- "This model is a custom Convolutional Neural Network trained from scratch."
 
64
  )
65
 
66
- st.sidebar.header("About this app")
 
 
 
67
  st.sidebar.markdown(
68
  """
69
- **Model:** CNN from scratch
70
- **Input size:** 224Γ—224
71
  **Classes:** 9 fish species
 
 
 
 
 
72
  **Author:** Enes Yamac
73
  """
74
  )
75
 
 
 
 
76
  uploaded_file = st.file_uploader(
77
- "Upload a fish image (JPG/PNG)",
78
  type=["jpg", "jpeg", "png"]
79
  )
80
 
 
 
 
81
  if uploaded_file is not None:
82
  image = Image.open(uploaded_file)
83
- st.image(image, caption="Uploaded image", use_column_width=True)
84
 
85
- with st.spinner("Predicting..."):
86
  pred_class, probs = predict_image(image)
87
 
88
- st.subheader("Prediction")
89
- st.write(f"**Predicted class:** {pred_class}")
90
 
91
- # Show probabilities for all classes
92
- st.subheader("Class probabilities")
93
  prob_dict = {cls: float(p) for cls, p in zip(CLASS_NAMES, probs)}
 
 
 
 
 
 
 
94
 
95
- st.bar_chart(prob_dict)
 
96
 
97
  else:
98
  st.info("Please upload an image to start.")
 
1
+ # --- Fish Classifier (CNN from scratch) ---------------------------------------
2
  import streamlit as st
3
  import numpy as np
4
+ import pandas as pd
5
  from PIL import Image
6
  import tensorflow as tf
7
  from tensorflow.keras.models import load_model
8
 
9
+ # -----------------------------------------------------------------------------
10
+ # PAGE CONFIG
11
+ # -----------------------------------------------------------------------------
12
  st.set_page_config(
13
+ page_title="Fish Classifier (CNN – From Scratch)",
14
  layout="centered"
15
  )
16
 
17
+ # -----------------------------------------------------------------------------
18
+ # SETTINGS
19
+ # -----------------------------------------------------------------------------
20
+ MODEL_PATH = "src/best_fish_cnn.keras"
21
  IMG_SIZE = (224, 224)
22
 
23
  CLASS_NAMES = [
 
32
  "Trout"
33
  ]
34
 
35
+ # -----------------------------------------------------------------------------
36
+ # MODEL LOADING (CACHED)
37
+ # -----------------------------------------------------------------------------
38
  @st.cache_resource
39
  def load_model_cached():
40
+ # compile=False reicht fΓΌr reine Vorhersage
41
+ model = load_model(MODEL_PATH, compile=False)
42
+ return model
43
 
44
  model = load_model_cached()
45
 
46
+ # -----------------------------------------------------------------------------
47
+ # HELPER FUNCTIONS
48
+ # -----------------------------------------------------------------------------
49
  def preprocess_image(image: Image.Image) -> np.ndarray:
50
  """Resize and normalize image for the CNN model."""
51
  image = image.convert("RGB")
52
  image = image.resize(IMG_SIZE)
53
+ img_array = np.array(image).astype("float32") / 255.0
54
+ img_array = np.expand_dims(img_array, axis=0)
55
  return img_array
56
 
57
  def predict_image(image: Image.Image):
58
+ """Run model prediction and return class label and probability vector."""
59
  img_batch = preprocess_image(image)
60
  preds = model.predict(img_batch)
61
+ probs = preds[0]
62
  pred_index = int(np.argmax(probs))
63
  pred_class = CLASS_NAMES[pred_index]
64
  return pred_class, probs
65
 
66
+ # -----------------------------------------------------------------------------
67
+ # UI – HEADER
68
+ # -----------------------------------------------------------------------------
 
 
 
69
  st.title("🐟 Fish Species Classifier (CNN From Scratch)")
70
+
71
  st.write(
72
  "Upload a fish image and the model will predict one of **9 fish species**. "
73
+ "This model is a custom Convolutional Neural Network (**no transfer learning**) "
74
+ "trained on the Kaggle fish dataset."
75
  )
76
 
77
+ # -----------------------------------------------------------------------------
78
+ # SIDEBAR INFO
79
+ # -----------------------------------------------------------------------------
80
+ st.sidebar.header("ℹ️ About this app")
81
  st.sidebar.markdown(
82
  """
83
+ **Model:** Custom CNN (no transfer learning)
84
+ **Input size:** 224 Γ— 224
85
  **Classes:** 9 fish species
86
+
87
+ **Validation performance:**
88
+ - Accuracy: ~84%
89
+ - Top-3 accuracy: ~95%
90
+
91
  **Author:** Enes Yamac
92
  """
93
  )
94
 
95
+ # -----------------------------------------------------------------------------
96
+ # FILE UPLOAD
97
+ # -----------------------------------------------------------------------------
98
  uploaded_file = st.file_uploader(
99
+ "πŸ“€ Upload a fish image (JPG/PNG)",
100
  type=["jpg", "jpeg", "png"]
101
  )
102
 
103
+ # -----------------------------------------------------------------------------
104
+ # PREDICTION + OUTPUT
105
+ # -----------------------------------------------------------------------------
106
  if uploaded_file is not None:
107
  image = Image.open(uploaded_file)
108
+ st.image(image, caption="Uploaded image", use_container_width=True)
109
 
110
+ with st.spinner("πŸ” Predicting..."):
111
  pred_class, probs = predict_image(image)
112
 
113
+ st.subheader("πŸ”Ž Prediction")
114
+ st.write(f"**Predicted class:** `{pred_class}`")
115
 
116
+ # Wahrscheinlichkeiten vorbereiten
 
117
  prob_dict = {cls: float(p) for cls, p in zip(CLASS_NAMES, probs)}
118
+ prob_df = (
119
+ pd.DataFrame.from_dict(prob_dict, orient="index", columns=["Probability"])
120
+ .sort_values("Probability", ascending=False)
121
+ )
122
+
123
+ st.subheader("πŸ“Š Class probabilities")
124
+ st.bar_chart(prob_df)
125
 
126
+ with st.expander("Show numeric probabilities"):
127
+ st.dataframe(prob_df.style.format("{:.2%}"))
128
 
129
  else:
130
  st.info("Please upload an image to start.")