Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -6,6 +6,27 @@ from tensorflow.keras.applications.vgg16 import VGG16, preprocess_input
|
|
6 |
from tensorflow.keras.preprocessing.image import img_to_array
|
7 |
from tensorflow.keras.preprocessing.sequence import pad_sequences
|
8 |
import pickle
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
# Custom function to handle attention mechanism
|
11 |
def attention_function(inputs):
|
@@ -70,53 +91,111 @@ def get_custom_objects(attention_func, output_shape_func):
|
|
70 |
|
71 |
# Multiple loading strategies with different attention mechanisms
|
72 |
def load_model_safely():
|
73 |
-
|
74 |
-
(attention_function, attention_output_shape),
|
75 |
-
(attention_function_v2, attention_output_shape),
|
76 |
-
(attention_function_v3, attention_output_shape),
|
77 |
-
]
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
|
90 |
-
#
|
91 |
try:
|
92 |
-
print("
|
93 |
model = tf.keras.models.load_model("caption_model.h5", compile=False)
|
94 |
-
print("
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
95 |
return model
|
|
|
96 |
except Exception as e:
|
97 |
-
print(f"
|
98 |
|
99 |
-
#
|
100 |
try:
|
101 |
-
print("
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
|
|
|
|
106 |
|
107 |
-
|
|
|
108 |
|
109 |
# Load your pre-trained model and tokenizer
|
110 |
-
|
111 |
-
|
112 |
-
except Exception as e:
|
113 |
-
print(f"Failed to load model: {e}")
|
114 |
-
print("Creating a dummy model for testing...")
|
115 |
-
# Create a simple dummy model for testing the interface
|
116 |
model = None
|
117 |
-
|
118 |
-
|
119 |
-
tokenizer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
|
121 |
# Image feature extractor model
|
122 |
feature_extractor = VGG16()
|
@@ -126,7 +205,10 @@ feature_extractor = tf.keras.Model(feature_extractor.input, feature_extractor.la
|
|
126 |
def generate_caption(image):
|
127 |
try:
|
128 |
if model is None:
|
129 |
-
return "Model failed to load. Please check the model file."
|
|
|
|
|
|
|
130 |
|
131 |
# Preprocess the image
|
132 |
image = image.resize((224, 224))
|
@@ -135,22 +217,27 @@ def generate_caption(image):
|
|
135 |
image = preprocess_input(image)
|
136 |
|
137 |
# Extract features
|
|
|
138 |
feature = feature_extractor.predict(image, verbose=0)
|
|
|
139 |
|
140 |
# Generate caption
|
141 |
input_text = 'startseq'
|
142 |
max_length = 34 # set this to your model's max_length
|
143 |
|
144 |
-
|
|
|
145 |
sequence = tokenizer.texts_to_sequences([input_text])[0]
|
146 |
sequence = pad_sequences([sequence], maxlen=max_length)
|
147 |
|
148 |
try:
|
|
|
149 |
yhat = model.predict([feature, sequence], verbose=0)
|
150 |
yhat = np.argmax(yhat)
|
|
|
151 |
except Exception as e:
|
152 |
-
print(f"Prediction error: {e}")
|
153 |
-
return f"Error during prediction: {str(e)}"
|
154 |
|
155 |
word = ''
|
156 |
for w, i in tokenizer.word_index.items():
|
@@ -158,15 +245,19 @@ def generate_caption(image):
|
|
158 |
word = w
|
159 |
break
|
160 |
|
|
|
161 |
if word == 'endseq' or word == '':
|
162 |
break
|
163 |
input_text += ' ' + word
|
164 |
|
165 |
caption = input_text.replace('startseq', '').strip()
|
166 |
-
|
|
|
167 |
|
168 |
except Exception as e:
|
169 |
-
|
|
|
|
|
170 |
|
171 |
# Gradio Interface
|
172 |
title = "πΈ Image Caption Generator"
|
|
|
6 |
from tensorflow.keras.preprocessing.image import img_to_array
|
7 |
from tensorflow.keras.preprocessing.sequence import pad_sequences
|
8 |
import pickle
|
9 |
+
import os
|
10 |
+
|
11 |
+
# Check if required files exist
|
12 |
+
def check_required_files():
|
13 |
+
required_files = ["caption_model.h5", "tokenizer.pkl"]
|
14 |
+
missing_files = []
|
15 |
+
|
16 |
+
for file in required_files:
|
17 |
+
if not os.path.exists(file):
|
18 |
+
missing_files.append(file)
|
19 |
+
else:
|
20 |
+
size = os.path.getsize(file)
|
21 |
+
print(f"β Found {file} ({size} bytes)")
|
22 |
+
|
23 |
+
if missing_files:
|
24 |
+
print(f"β Missing files: {missing_files}")
|
25 |
+
return False
|
26 |
+
return True
|
27 |
+
|
28 |
+
print("Checking required files...")
|
29 |
+
files_exist = check_required_files()
|
30 |
|
31 |
# Custom function to handle attention mechanism
|
32 |
def attention_function(inputs):
|
|
|
91 |
|
92 |
# Multiple loading strategies with different attention mechanisms
|
93 |
def load_model_safely():
|
94 |
+
print("Starting model loading process...")
|
|
|
|
|
|
|
|
|
95 |
|
96 |
+
# Strategy 1: Try with custom Lambda that handles the attention operation
|
97 |
+
try:
|
98 |
+
print("Strategy 1: Loading with custom attention Lambda...")
|
99 |
+
|
100 |
+
def custom_attention(inputs):
|
101 |
+
"""Handle attention mechanism between two inputs"""
|
102 |
+
if len(inputs) == 2:
|
103 |
+
attention_weights, features = inputs
|
104 |
+
# Simple attention: multiply attention weights with features
|
105 |
+
# Expand attention weights to match feature dimensions
|
106 |
+
if len(attention_weights.shape) == 3 and len(features.shape) == 3:
|
107 |
+
attention_weights = tf.expand_dims(attention_weights, axis=-1)
|
108 |
+
return tf.multiply(attention_weights, features)
|
109 |
+
return inputs[0] if isinstance(inputs, list) else inputs
|
110 |
+
|
111 |
+
custom_objects = {
|
112 |
+
'Lambda': lambda function=None, output_shape=None, **kwargs:
|
113 |
+
tf.keras.layers.Lambda(
|
114 |
+
custom_attention if function is None else function,
|
115 |
+
output_shape=lambda input_shape: input_shape[1] if isinstance(input_shape, list) else input_shape,
|
116 |
+
**kwargs
|
117 |
+
)
|
118 |
+
}
|
119 |
+
|
120 |
+
model = tf.keras.models.load_model("caption_model.h5", custom_objects=custom_objects)
|
121 |
+
print("β Strategy 1 successful!")
|
122 |
+
return model
|
123 |
+
|
124 |
+
except Exception as e:
|
125 |
+
print(f"β Strategy 1 failed: {str(e)[:200]}...")
|
126 |
|
127 |
+
# Strategy 2: Load with compile=False and try to fix compilation later
|
128 |
try:
|
129 |
+
print("Strategy 2: Loading without compilation...")
|
130 |
model = tf.keras.models.load_model("caption_model.h5", compile=False)
|
131 |
+
print("β Strategy 2 successful!")
|
132 |
+
return model
|
133 |
+
|
134 |
+
except Exception as e:
|
135 |
+
print(f"β Strategy 2 failed: {str(e)[:200]}...")
|
136 |
+
|
137 |
+
# Strategy 3: Try loading with TensorFlow's built-in Lambda handling
|
138 |
+
try:
|
139 |
+
print("Strategy 3: Loading with default Lambda handling...")
|
140 |
+
|
141 |
+
def identity_function(x):
|
142 |
+
if isinstance(x, list) and len(x) == 2:
|
143 |
+
# For attention mechanism, return the second input (features)
|
144 |
+
return x[1]
|
145 |
+
return x
|
146 |
+
|
147 |
+
custom_objects = {
|
148 |
+
'Lambda': lambda function=identity_function, output_shape=None, **kwargs:
|
149 |
+
tf.keras.layers.Lambda(
|
150 |
+
function,
|
151 |
+
output_shape=lambda input_shape: input_shape[1] if isinstance(input_shape, list) else input_shape,
|
152 |
+
**kwargs
|
153 |
+
)
|
154 |
+
}
|
155 |
+
|
156 |
+
model = tf.keras.models.load_model("caption_model.h5", custom_objects=custom_objects)
|
157 |
+
print("β Strategy 3 successful!")
|
158 |
return model
|
159 |
+
|
160 |
except Exception as e:
|
161 |
+
print(f"β Strategy 3 failed: {str(e)[:200]}...")
|
162 |
|
163 |
+
# Strategy 4: Try with minimal custom objects
|
164 |
try:
|
165 |
+
print("Strategy 4: Loading with minimal custom objects...")
|
166 |
+
model = tf.keras.models.load_model("caption_model.h5", custom_objects={'Lambda': tf.keras.layers.Lambda})
|
167 |
+
print("β Strategy 4 successful!")
|
168 |
+
return model
|
169 |
+
|
170 |
+
except Exception as e:
|
171 |
+
print(f"β Strategy 4 failed: {str(e)[:200]}...")
|
172 |
|
173 |
+
print("All strategies failed. Model could not be loaded.")
|
174 |
+
raise Exception("All model loading strategies failed. The model file may be corrupted or incompatible.")
|
175 |
|
176 |
# Load your pre-trained model and tokenizer
|
177 |
+
if not files_exist:
|
178 |
+
print("Cannot proceed without required files.")
|
|
|
|
|
|
|
|
|
179 |
model = None
|
180 |
+
tokenizer = None
|
181 |
+
else:
|
182 |
+
# Load tokenizer first
|
183 |
+
try:
|
184 |
+
with open("tokenizer.pkl", "rb") as handle:
|
185 |
+
tokenizer = pickle.load(handle)
|
186 |
+
print("β Tokenizer loaded successfully")
|
187 |
+
except Exception as e:
|
188 |
+
print(f"β Failed to load tokenizer: {e}")
|
189 |
+
tokenizer = None
|
190 |
+
|
191 |
+
# Load model
|
192 |
+
try:
|
193 |
+
model = load_model_safely()
|
194 |
+
print("β Model loaded successfully and ready for inference!")
|
195 |
+
except Exception as e:
|
196 |
+
print(f"β Failed to load model: {e}")
|
197 |
+
print("The app will not work without a properly loaded model.")
|
198 |
+
model = None
|
199 |
|
200 |
# Image feature extractor model
|
201 |
feature_extractor = VGG16()
|
|
|
205 |
def generate_caption(image):
|
206 |
try:
|
207 |
if model is None:
|
208 |
+
return "β Model failed to load. Please check the model file and console output for details."
|
209 |
+
|
210 |
+
if tokenizer is None:
|
211 |
+
return "β Tokenizer failed to load. Please check the tokenizer.pkl file."
|
212 |
|
213 |
# Preprocess the image
|
214 |
image = image.resize((224, 224))
|
|
|
217 |
image = preprocess_input(image)
|
218 |
|
219 |
# Extract features
|
220 |
+
print("Extracting image features...")
|
221 |
feature = feature_extractor.predict(image, verbose=0)
|
222 |
+
print(f"Features extracted, shape: {feature.shape}")
|
223 |
|
224 |
# Generate caption
|
225 |
input_text = 'startseq'
|
226 |
max_length = 34 # set this to your model's max_length
|
227 |
|
228 |
+
print("Starting caption generation...")
|
229 |
+
for i in range(max_length):
|
230 |
sequence = tokenizer.texts_to_sequences([input_text])[0]
|
231 |
sequence = pad_sequences([sequence], maxlen=max_length)
|
232 |
|
233 |
try:
|
234 |
+
print(f"Prediction step {i+1}: input_text = '{input_text}'")
|
235 |
yhat = model.predict([feature, sequence], verbose=0)
|
236 |
yhat = np.argmax(yhat)
|
237 |
+
print(f"Predicted token index: {yhat}")
|
238 |
except Exception as e:
|
239 |
+
print(f"Prediction error at step {i+1}: {e}")
|
240 |
+
return f"β Error during prediction: {str(e)}"
|
241 |
|
242 |
word = ''
|
243 |
for w, i in tokenizer.word_index.items():
|
|
|
245 |
word = w
|
246 |
break
|
247 |
|
248 |
+
print(f"Predicted word: '{word}'")
|
249 |
if word == 'endseq' or word == '':
|
250 |
break
|
251 |
input_text += ' ' + word
|
252 |
|
253 |
caption = input_text.replace('startseq', '').strip()
|
254 |
+
print(f"Final caption: '{caption}'")
|
255 |
+
return f"β
{caption}" if caption else "β Unable to generate caption"
|
256 |
|
257 |
except Exception as e:
|
258 |
+
error_msg = f"β Error processing image: {str(e)}"
|
259 |
+
print(error_msg)
|
260 |
+
return error_msg
|
261 |
|
262 |
# Gradio Interface
|
263 |
title = "πΈ Image Caption Generator"
|