BunkerMinecraft commited on
Commit
436fc42
·
verified ·
1 Parent(s): d1ba352

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +28 -40
app.py CHANGED
@@ -1,23 +1,21 @@
1
  from huggingface_hub import hf_hub_download
2
  import tensorflow as tf
3
  from tensorflow.keras import layers, models, regularizers
4
- import json
 
5
  from PIL import Image
6
- import numpy as np
7
- import os
8
  import gradio as gr
9
 
10
- # ---- Load config if present ----
11
- cfg = None
12
- if os.path.exists("config.json"):
13
- with open("config.json") as f:
14
- cfg = json.load(f)
15
-
16
  # ---- Class labels ----
17
- class_names = ['battery', 'glass', 'metal', 'organic', 'paper', 'plastic']
 
18
 
19
- # ---- Build model architecture ----
20
- base_model = tf.keras.applications.EfficientNetB7(weights=None)
 
 
 
 
21
  model = models.Sequential([
22
  base_model,
23
  layers.GlobalAveragePooling2D(),
@@ -25,18 +23,17 @@ model = models.Sequential([
25
  layers.Dropout(0.3),
26
  layers.Dense(128, activation='relu', kernel_regularizer=regularizers.l2(0.001)),
27
  layers.Dropout(0.3),
28
- layers.Dense(6, activation='softmax')
29
  ])
30
 
31
  # ---- Load trained weights (pointer-safe) ----
32
  WEIGHTS_LOCAL = "model_weights.h5"
33
- MODEL_REPO_ID = "BunkerMinecraft/neurecaiv2" # change to your model repo
34
 
35
  def _is_pointer(path: str) -> bool:
36
  if not os.path.exists(path) or os.path.getsize(path) < 1024:
37
  try:
38
- with open(path, "rb") as f:
39
- head = f.read(200)
40
  return (b"git-lfs" in head) or (b"xet" in head)
41
  except Exception:
42
  return True
@@ -48,41 +45,32 @@ if _is_pointer(WEIGHTS_LOCAL):
48
  filename="model_weights.h5",
49
  repo_type="model"
50
  )
51
-
52
  model.load_weights(WEIGHTS_LOCAL)
53
  print(f"Loaded weights from: {WEIGHTS_LOCAL}")
54
 
 
 
 
 
 
 
55
 
56
- # ---- Preprocess image ----
57
- def preprocess(image: Image.Image):
58
- image = image.resize((224, 224))
59
- image = np.array(image) / 255.0
60
- image = np.expand_dims(image, axis=0)
61
- return image
62
-
63
- # ---- Prediction function ----
64
  def predict(image: Image.Image):
65
  x = preprocess(image)
66
- preds = model.predict(x)
67
- class_idx = int(np.argmax(preds, axis=1)[0])
68
- confidence = float(np.max(preds))
69
- return {
70
- "class": class_names[class_idx],
71
- "confidence": confidence
72
- }
73
 
74
- # ---- Gradio interface ----
75
  interface = gr.Interface(
76
  fn=predict,
77
  inputs=gr.Image(type="pil", label="Upload an image"),
78
- outputs=[
79
- gr.Label(num_top_classes=1, label="Predicted class"),
80
- gr.Number(label="Confidence")
81
- ],
82
  title="NeuRecAI - Material Classifier",
83
- description="Upload an image to classify it as battery, glass, metal, organic, paper, or plastic."
84
  )
85
 
86
- # ---- Launch ----
87
  if __name__ == "__main__":
88
- interface.launch()
 
1
  from huggingface_hub import hf_hub_download
2
  import tensorflow as tf
3
  from tensorflow.keras import layers, models, regularizers
4
+ from tensorflow.keras.applications.efficientnet import preprocess_input
5
+ import json, os, numpy as np
6
  from PIL import Image
 
 
7
  import gradio as gr
8
 
 
 
 
 
 
 
9
  # ---- Class labels ----
10
+ class_names = ['battery','glass','metal','organic','paper','plastic']
11
+ num_classes = len(class_names)
12
 
13
+ # ---- Build model (match training) ----
14
+ base_model = tf.keras.applications.EfficientNetB7(
15
+ weights=None, # or "imagenet" if that’s how you trained
16
+ include_top=False,
17
+ input_shape=(224, 224, 3)
18
+ )
19
  model = models.Sequential([
20
  base_model,
21
  layers.GlobalAveragePooling2D(),
 
23
  layers.Dropout(0.3),
24
  layers.Dense(128, activation='relu', kernel_regularizer=regularizers.l2(0.001)),
25
  layers.Dropout(0.3),
26
+ layers.Dense(num_classes, activation='softmax')
27
  ])
28
 
29
  # ---- Load trained weights (pointer-safe) ----
30
  WEIGHTS_LOCAL = "model_weights.h5"
31
+ MODEL_REPO_ID = "BunkerMinecraft/neurecaiv2" # your model repo
32
 
33
  def _is_pointer(path: str) -> bool:
34
  if not os.path.exists(path) or os.path.getsize(path) < 1024:
35
  try:
36
+ with open(path, "rb") as f: head = f.read(200)
 
37
  return (b"git-lfs" in head) or (b"xet" in head)
38
  except Exception:
39
  return True
 
45
  filename="model_weights.h5",
46
  repo_type="model"
47
  )
 
48
  model.load_weights(WEIGHTS_LOCAL)
49
  print(f"Loaded weights from: {WEIGHTS_LOCAL}")
50
 
51
+ # ---- Preprocess ----
52
+ def preprocess(img: Image.Image):
53
+ img = img.convert("RGB").resize((224, 224), Image.BILINEAR)
54
+ x = np.asarray(img, dtype=np.float32)
55
+ x = preprocess_input(x) # EfficientNet normalization
56
+ return np.expand_dims(x, 0)
57
 
58
+ # ---- Predict ----
 
 
 
 
 
 
 
59
  def predict(image: Image.Image):
60
  x = preprocess(image)
61
+ y = model.predict(x, verbose=0)[0]
62
+ idx = int(np.argmax(y))
63
+ conf = float(y[idx])
64
+ return class_names[idx], conf
 
 
 
65
 
66
+ # ---- UI ----
67
  interface = gr.Interface(
68
  fn=predict,
69
  inputs=gr.Image(type="pil", label="Upload an image"),
70
+ outputs=[gr.Label(label="Predicted class"), gr.Number(label="Confidence")],
 
 
 
71
  title="NeuRecAI - Material Classifier",
72
+ description="Battery, glass, metal, organic, paper, or plastic."
73
  )
74
 
 
75
  if __name__ == "__main__":
76
+ interface.launch()