Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -3,14 +3,46 @@ import streamlit as st
|
|
3 |
from streamlit.runtime.uploaded_file_manager import UploadedFile
|
4 |
import tensorflow as tf
|
5 |
import pandas as pd
|
6 |
-
from PIL import Image
|
7 |
|
8 |
# 🔹 Expand the Page Layout
|
9 |
-
st.set_page_config(layout="wide")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
# --- Constants and Data ---
|
12 |
current_model = "Model Mini"
|
13 |
-
new_model = "Food Vision"
|
|
|
14 |
class_names = ['apple_pie', 'baby_back_ribs', 'baklava', 'beef_carpaccio', 'beef_tartare',
|
15 |
'beet_salad', 'beignets', 'bibimbap', 'bread_pudding', 'breakfast_burrito',
|
16 |
'bruschetta', 'caesar_salad', 'cannoli', 'caprese_salad', 'carrot_cake',
|
@@ -33,7 +65,7 @@ class_names = ['apple_pie', 'baby_back_ribs', 'baklava', 'beef_carpaccio', 'beef
|
|
33 |
'sushi', 'tacos', 'takoyaki', 'tiramisu', 'tuna_tartare', 'waffles']
|
34 |
|
35 |
top_ten_dict = {
|
36 |
-
"class_name": ["edamame", "macarons", "oysters", "pho", "mussels",
|
37 |
"sashimi", "seaweed_salad", "dumplings", "guacamole", "onion_rings"],
|
38 |
"f1-score": [0.964427, 0.900433, 0.853119, 0.852652, 0.850622,
|
39 |
0.844794, 0.834356, 0.833006, 0.83209, 0.831967]
|
@@ -45,72 +77,100 @@ last_ten_dict = {
|
|
45 |
0.340426, 0.340045, 0.339785, 0.324826, 0.282407]
|
46 |
}
|
47 |
|
48 |
-
# 🔹 Custom CSS for
|
49 |
st.markdown(
|
50 |
"""
|
51 |
<style>
|
52 |
-
/* Center content
|
53 |
.centered {
|
54 |
display: flex;
|
55 |
flex-direction: column;
|
56 |
-
align-items: center;
|
57 |
-
justify-content:
|
58 |
text-align: center;
|
59 |
-
width: 100%;
|
60 |
-
min-height:
|
61 |
-
padding-top:
|
62 |
-
padding-bottom:
|
|
|
63 |
}
|
64 |
|
65 |
-
/*
|
66 |
-
/*
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
div[data-testid="stFileUploader"] > section {
|
68 |
-
|
|
|
|
|
69 |
}
|
70 |
-
div[data-testid="stFileUploader"]
|
71 |
-
|
|
|
|
|
|
|
72 |
}
|
73 |
-
div[data-testid="stFileUploader"] label {
|
74 |
-
/* Style the label if needed */
|
75 |
-
}
|
76 |
|
77 |
|
78 |
/* Center images and standardize size */
|
79 |
-
.centered img {
|
80 |
display: block;
|
81 |
margin-left: auto;
|
82 |
margin-right: auto;
|
83 |
-
max-width: 200px;
|
84 |
-
max-height: 200px;
|
85 |
-
width: auto;
|
86 |
-
height: auto;
|
87 |
-
object-fit: contain;
|
88 |
border-radius: 20px;
|
89 |
-
margin-bottom:
|
90 |
}
|
91 |
|
92 |
-
/*
|
93 |
div[data-testid="stVerticalBlock"] div[data-testid="stHorizontalBlock"] {
|
94 |
-
align-items:
|
95 |
}
|
96 |
|
97 |
/* Style the radio buttons */
|
98 |
div[data-testid="stRadio"] > label {
|
99 |
-
font-weight: bold;
|
100 |
-
margin-bottom:
|
|
|
101 |
}
|
102 |
div[data-testid="stRadio"] > div {
|
103 |
-
|
104 |
-
|
105 |
-
|
|
|
106 |
}
|
107 |
|
108 |
/* Style the button */
|
109 |
div[data-testid="stButton"] > button {
|
110 |
-
width: 80%;
|
111 |
-
margin-top:
|
|
|
|
|
|
|
|
|
|
|
|
|
112 |
}
|
113 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
114 |
</style>
|
115 |
""",
|
116 |
unsafe_allow_html=True
|
@@ -122,6 +182,7 @@ st.header("A food vision app using a CNN model fine-tuned on EfficientNet.")
|
|
122 |
st.divider()
|
123 |
|
124 |
# --- Explanations (Collapsible) ---
|
|
|
125 |
with st.expander("Learn More: What is a CNN?"):
|
126 |
st.write("""
|
127 |
A Neural Network is a system inspired by the human brain, composed of interconnected nodes (neurons) organized in layers: an input layer, one or more hidden layers, and an output layer.
|
@@ -310,36 +371,27 @@ with st.expander("What is the F1-Score?"):
|
|
310 |
st.subheader("Top and Least Performing Classes (by F1-Score)")
|
311 |
with st.container():
|
312 |
top_ten_df = pd.DataFrame(top_ten_dict).sort_values("f1-score", ascending=False)
|
313 |
-
last_ten_df = pd.DataFrame(last_ten_dict).sort_values("f1-score", ascending=True)
|
314 |
|
315 |
-
# Format class names for display
|
316 |
top_ten_df['class_name_display'] = top_ten_df['class_name'].str.replace('_', ' ').str.title()
|
317 |
last_ten_df['class_name_display'] = last_ten_df['class_name'].str.replace('_', ' ').str.title()
|
318 |
|
319 |
-
|
320 |
col1, col2 = st.columns(2)
|
321 |
with col1:
|
322 |
st.write("**Top 10 Classes**")
|
323 |
-
st.bar_chart(top_ten_df.set_index('class_name_display')['f1-score'],
|
324 |
-
# horizontal=True, # Bar chart auto-detects horizontal best here
|
325 |
-
use_container_width=True)
|
326 |
with col2:
|
327 |
st.write("**Bottom 10 Classes**")
|
328 |
-
st.bar_chart(last_ten_df.set_index('class_name_display')['f1-score'],
|
329 |
-
|
330 |
-
use_container_width=True, color="#ff748c") # Red color for low scores
|
331 |
-
st.divider()
|
332 |
-
|
333 |
|
334 |
# --- Helper Functions ---
|
335 |
-
@st.cache_resource
|
336 |
def load_model(filepath):
|
337 |
"""Loads a Tensorflow Keras Model."""
|
338 |
-
st.write(f"Cache miss: Loading model from {filepath}")
|
339 |
try:
|
340 |
model = tf.keras.models.load_model(filepath)
|
341 |
-
# You might need a warm-up prediction for GPU memory allocation
|
342 |
-
# For example: model.predict(tf.zeros([1, 224, 224, 3]))
|
343 |
return model
|
344 |
except Exception as e:
|
345 |
st.error(f"Error loading model from {filepath}: {e}")
|
@@ -348,21 +400,10 @@ def load_model(filepath):
|
|
348 |
def load_prep_image(image_input: UploadedFile, img_shape=224):
|
349 |
"""Reads and preprocesses an image for EfficientNet prediction."""
|
350 |
try:
|
351 |
-
# Read image file buffer
|
352 |
bytes_data = image_input.getvalue()
|
353 |
-
# Decode image
|
354 |
image_tensor = tf.io.decode_image(bytes_data, channels=3)
|
355 |
-
# Resize image
|
356 |
-
# Use tf.image.resize with method='nearest' or 'bilinear' (default)
|
357 |
image_tensor_resized = tf.image.resize(image_tensor, [img_shape, img_shape])
|
358 |
-
# Expand dimensions to create batch_size 1 -> (1, H, W, C)
|
359 |
image_tensor_expanded = tf.expand_dims(image_tensor_resized, axis=0)
|
360 |
-
# EfficientNet models usually have their own preprocessing layer/function
|
361 |
-
# or expect inputs scaled 0-255. Check the specific model's requirement.
|
362 |
-
# If it expects 0-1 scaling and doesn't do it internally:
|
363 |
-
# image_tensor_scaled = image_tensor_expanded / 255.0
|
364 |
-
# return image_tensor_scaled
|
365 |
-
# Assuming EfficientNet B0 handles scaling or expects 0-255:
|
366 |
return image_tensor_expanded
|
367 |
except Exception as e:
|
368 |
st.error(f"Error processing image: {e}")
|
@@ -385,19 +426,20 @@ def predict_using_model(image_input: UploadedFile, model_path: str) -> tuple[str
|
|
385 |
try:
|
386 |
with st.spinner("🤖 Model is predicting..."):
|
387 |
pred_prob = model.predict(processed_image)
|
388 |
-
predicted_index = tf.argmax(pred_prob, axis=1).numpy()[0]
|
389 |
predicted_class_name = class_names[predicted_index]
|
390 |
-
predicted_probability = float(tf.reduce_max(pred_prob).numpy())
|
391 |
return predicted_class_name, predicted_probability
|
392 |
except Exception as e:
|
393 |
st.error(f"Prediction failed: {e}")
|
394 |
return None, None
|
395 |
|
396 |
# --- Interactive Demo Section ---
|
397 |
-
|
398 |
st.header(f"Try the Models: :blue[{current_model}] & :blue[{new_model}]")
|
399 |
st.caption("_Model performance may vary. Models are periodically updated._")
|
400 |
|
|
|
401 |
# Initialize session state keys if they don't exist
|
402 |
if "prediction_result" not in st.session_state:
|
403 |
st.session_state.prediction_result = None
|
@@ -408,18 +450,19 @@ if "predicted_prob" not in st.session_state:
|
|
408 |
|
409 |
|
410 |
# Use columns for layout
|
411 |
-
cols = st.columns([3, 0.5, 2, 0.5, 3], gap="medium") #
|
|
|
412 |
|
413 |
# --- Column 1: Image Input ---
|
414 |
with cols[0]:
|
415 |
st.markdown('<div class="centered">', unsafe_allow_html=True) # Apply centering
|
416 |
-
st.subheader("1. Provide an Image")
|
417 |
image_source = st.radio(
|
418 |
"Choose image source:",
|
419 |
("Upload Image", "Use Camera"),
|
420 |
key="image_source",
|
421 |
horizontal=True,
|
422 |
-
label_visibility="collapsed"
|
423 |
)
|
424 |
|
425 |
uploaded_image = None
|
@@ -442,22 +485,25 @@ with cols[0]:
|
|
442 |
|
443 |
# Display uploaded image preview
|
444 |
if uploaded_image:
|
445 |
-
image_bytes_for_state = uploaded_image.getvalue()
|
446 |
-
st.image(image_bytes_for_state, caption="Your image", use_column_width='auto')
|
447 |
-
|
448 |
else:
|
449 |
st.info("Upload or take a picture.")
|
450 |
|
451 |
-
st.markdown('</div>', unsafe_allow_html=True)
|
|
|
452 |
|
453 |
# --- Column 2: Arrow 1 ---
|
454 |
with cols[1]:
|
455 |
-
|
|
|
|
|
456 |
|
457 |
# --- Column 3: Model Selection & Prediction ---
|
458 |
with cols[2]:
|
459 |
-
st.markdown('<div class="centered">', unsafe_allow_html=True)
|
460 |
-
st.subheader("2. Select Model")
|
461 |
|
462 |
chosen_model = st.radio(
|
463 |
"Pick a Model:",
|
@@ -470,17 +516,16 @@ with cols[2]:
|
|
470 |
model_path_to_use = ""
|
471 |
model_image_path = ""
|
472 |
|
473 |
-
if chosen_model == current_model:
|
474 |
-
model_image_path = "
|
475 |
-
model_path_to_use = "model_mini_Food101.keras"
|
476 |
-
elif chosen_model == new_model:
|
477 |
-
model_image_path = "content/creativity_15557951.png"
|
478 |
-
model_path_to_use = "FoodVision.keras"
|
479 |
|
480 |
-
# Display model icon/image if path is valid
|
481 |
try:
|
482 |
if model_image_path:
|
483 |
-
st.image(model_image_path, width=150) #
|
484 |
except Exception as e:
|
485 |
st.warning(f"Could not load model image: {model_image_path}")
|
486 |
|
@@ -489,41 +534,39 @@ with cols[2]:
|
|
489 |
label="Predict Food!",
|
490 |
icon="⚛️",
|
491 |
type="primary",
|
492 |
-
use_container_width=True,
|
493 |
-
disabled=not uploaded_image or not model_path_to_use
|
494 |
)
|
495 |
|
496 |
if predict_button:
|
497 |
if uploaded_image and model_path_to_use:
|
498 |
-
# Perform prediction
|
499 |
result_class, result_prob = predict_using_model(uploaded_image, model_path=model_path_to_use)
|
500 |
-
# Store results in session state
|
501 |
st.session_state.prediction_result = result_class
|
502 |
st.session_state.predicted_prob = result_prob
|
503 |
-
st.session_state.predicted_image_bytes = image_bytes_for_state
|
504 |
else:
|
505 |
st.warning("Please provide an image and select a valid model.")
|
506 |
|
507 |
-
st.markdown('</div>', unsafe_allow_html=True)
|
|
|
508 |
|
509 |
# --- Column 4: Arrow 2 ---
|
510 |
with cols[3]:
|
511 |
-
|
|
|
|
|
512 |
|
513 |
# --- Column 5: Output ---
|
514 |
with cols[4]:
|
515 |
-
st.markdown('<div class="centered">', unsafe_allow_html=True)
|
516 |
-
st.subheader("3. Prediction Result")
|
517 |
|
518 |
-
# Display result from session state
|
519 |
if st.session_state.prediction_result and st.session_state.predicted_image_bytes:
|
520 |
-
# Display the image associated with the prediction
|
521 |
st.image(st.session_state.predicted_image_bytes, caption="Image Analyzed", use_column_width='auto')
|
522 |
|
523 |
result_class = st.session_state.prediction_result
|
524 |
probability = st.session_state.predicted_prob
|
525 |
|
526 |
-
# Format class name nicely
|
527 |
if "_" in result_class:
|
528 |
modified_class = result_class.replace("_", " ").title()
|
529 |
else:
|
@@ -531,15 +574,15 @@ with cols[4]:
|
|
531 |
|
532 |
st.success(f"Prediction: **:blue[{modified_class}]**")
|
533 |
if probability:
|
534 |
-
st.write(f"Confidence: {probability:.
|
535 |
|
536 |
elif predict_button:
|
537 |
-
|
538 |
-
st.error("Prediction could not be completed. Check logs or try again.")
|
539 |
else:
|
540 |
-
st.info("Result will appear here
|
|
|
|
|
541 |
|
542 |
-
st.markdown('</div>', unsafe_allow_html=True) # Close centered div
|
543 |
|
544 |
# --- Footer or Final Divider ---
|
545 |
-
st.divider()
|
|
|
3 |
from streamlit.runtime.uploaded_file_manager import UploadedFile
|
4 |
import tensorflow as tf
|
5 |
import pandas as pd
|
6 |
+
from PIL import Image
|
7 |
|
8 |
# 🔹 Expand the Page Layout
|
9 |
+
st.set_page_config(layout="wide")
|
10 |
+
|
11 |
+
# 2. Inject CSS to override the default max-width and set it to 90%
|
12 |
+
st.markdown(
|
13 |
+
"""
|
14 |
+
<style>
|
15 |
+
/* Target the main block container */
|
16 |
+
.block-container {
|
17 |
+
/* Set the max-width to 90% of the viewport */
|
18 |
+
max-width: 90% !important;
|
19 |
+
|
20 |
+
/* Optional: Adjust padding if needed */
|
21 |
+
/* padding-left: 2rem !important; */
|
22 |
+
/* padding-right: 2rem !important; */
|
23 |
+
/* padding-top: 1rem !important; */ /* Adjust top padding */
|
24 |
+
/* padding-bottom: 1rem !important; */
|
25 |
+
|
26 |
+
/* Ensure it remains centered */
|
27 |
+
margin: auto !important;
|
28 |
+
}
|
29 |
+
|
30 |
+
/* You might need to target a more specific element depending on the Streamlit version */
|
31 |
+
/* If the above doesn't work, try inspecting the element in your browser */
|
32 |
+
/* and use a selector like: */
|
33 |
+
/* div[data-testid="stAppViewContainer"] > section > div[data-testid="stBlock"] { */
|
34 |
+
/* max-width: 90% !important; */
|
35 |
+
/* } */
|
36 |
+
|
37 |
+
</style>
|
38 |
+
""",
|
39 |
+
unsafe_allow_html=True
|
40 |
+
)
|
41 |
|
42 |
# --- Constants and Data ---
|
43 |
current_model = "Model Mini"
|
44 |
+
new_model = "Food Vision"
|
45 |
+
# ... (class_names, top_ten_dict, last_ten_dict remain the same) ...
|
46 |
class_names = ['apple_pie', 'baby_back_ribs', 'baklava', 'beef_carpaccio', 'beef_tartare',
|
47 |
'beet_salad', 'beignets', 'bibimbap', 'bread_pudding', 'breakfast_burrito',
|
48 |
'bruschetta', 'caesar_salad', 'cannoli', 'caprese_salad', 'carrot_cake',
|
|
|
65 |
'sushi', 'tacos', 'takoyaki', 'tiramisu', 'tuna_tartare', 'waffles']
|
66 |
|
67 |
top_ten_dict = {
|
68 |
+
"class_name": ["edamame", "macarons", "oysters", "pho", "mussels",
|
69 |
"sashimi", "seaweed_salad", "dumplings", "guacamole", "onion_rings"],
|
70 |
"f1-score": [0.964427, 0.900433, 0.853119, 0.852652, 0.850622,
|
71 |
0.844794, 0.834356, 0.833006, 0.83209, 0.831967]
|
|
|
77 |
0.340426, 0.340045, 0.339785, 0.324826, 0.282407]
|
78 |
}
|
79 |
|
80 |
+
# 🔹 Custom CSS for Alignment and Spacing
|
81 |
st.markdown(
|
82 |
"""
|
83 |
<style>
|
84 |
+
/* Center content H, align V top, reduce padding/min-height */
|
85 |
.centered {
|
86 |
display: flex;
|
87 |
flex-direction: column;
|
88 |
+
align-items: center; /* Keep horizontal centering */
|
89 |
+
justify-content: flex-start; /* Align content to the top */
|
90 |
text-align: center;
|
91 |
+
width: 100%;
|
92 |
+
min-height: 100px; /* Reduced minimum height */
|
93 |
+
padding-top: 5px; /* Reduced padding */
|
94 |
+
padding-bottom: 10px;
|
95 |
+
height: 100%; /* Allow div to take full column height if needed */
|
96 |
}
|
97 |
|
98 |
+
/* Ensure subheaders within columns have less top margin */
|
99 |
+
.centered h3 { /* Targets the subheaders like "1. Provide Image" */
|
100 |
+
margin-top: 0px !important;
|
101 |
+
padding-top: 0px !important;
|
102 |
+
margin-bottom: 15px !important; /* Add space below subheader */
|
103 |
+
}
|
104 |
+
|
105 |
+
/* Style file uploader (ensure consistent padding) */
|
106 |
div[data-testid="stFileUploader"] > section {
|
107 |
+
padding: 0;
|
108 |
+
display: flex; /* Use flex for centering button inside */
|
109 |
+
justify-content: center;
|
110 |
}
|
111 |
+
div[data-testid="stFileUploader"] label { /* Style the button-like label */
|
112 |
+
margin-bottom: 10px; /* Space below uploader */
|
113 |
+
}
|
114 |
+
div[data-testid="stCameraInput"] label { /* Style the camera input label */
|
115 |
+
margin-bottom: 10px; /* Space below camera input */
|
116 |
}
|
|
|
|
|
|
|
117 |
|
118 |
|
119 |
/* Center images and standardize size */
|
120 |
+
.centered img {
|
121 |
display: block;
|
122 |
margin-left: auto;
|
123 |
margin-right: auto;
|
124 |
+
max-width: 200px;
|
125 |
+
max-height: 200px;
|
126 |
+
width: auto;
|
127 |
+
height: auto;
|
128 |
+
object-fit: contain;
|
129 |
border-radius: 20px;
|
130 |
+
margin-bottom: 10px; /* Reduced space below image */
|
131 |
}
|
132 |
|
133 |
+
/* Align the TOPS of the columns */
|
134 |
div[data-testid="stVerticalBlock"] div[data-testid="stHorizontalBlock"] {
|
135 |
+
align-items: flex-start !important; /* Align column tops */
|
136 |
}
|
137 |
|
138 |
/* Style the radio buttons */
|
139 |
div[data-testid="stRadio"] > label {
|
140 |
+
font-weight: bold;
|
141 |
+
margin-bottom: 5px !important; /* Reduced margin */
|
142 |
+
padding-top: 0 !important;
|
143 |
}
|
144 |
div[data-testid="stRadio"] > div {
|
145 |
+
display: flex;
|
146 |
+
justify-content: center;
|
147 |
+
gap: 10px; /* Reduced gap */
|
148 |
+
margin-bottom: 10px; /* Add space below radio group */
|
149 |
}
|
150 |
|
151 |
/* Style the button */
|
152 |
div[data-testid="stButton"] > button {
|
153 |
+
width: 80%;
|
154 |
+
margin-top: 15px; /* Reduced margin */
|
155 |
+
}
|
156 |
+
|
157 |
+
/* Reduce space BELOW the caption ABOVE the columns */
|
158 |
+
div[data-testid="stCaptionContainer"] {
|
159 |
+
padding-bottom: 0px !important;
|
160 |
+
margin-bottom: -10px !important; /* Negative margin pulls following elements up */
|
161 |
}
|
162 |
|
163 |
+
/* Reduce space ABOVE the main H2 header for the demo section */
|
164 |
+
h2[data-testid="stHeading"]:has(+ div[data-testid="stCaptionContainer"]) {
|
165 |
+
margin-bottom: 5px !important; /* Space between header and caption */
|
166 |
+
}
|
167 |
+
|
168 |
+
/* Reduce space below the final divider before the demo H2 header */
|
169 |
+
hr[data-testid="stDivider"] + h2[data-testid="stHeading"] {
|
170 |
+
margin-top: -15px !important; /* Pull header closer to divider */
|
171 |
+
}
|
172 |
+
|
173 |
+
|
174 |
</style>
|
175 |
""",
|
176 |
unsafe_allow_html=True
|
|
|
182 |
st.divider()
|
183 |
|
184 |
# --- Explanations (Collapsible) ---
|
185 |
+
# ... (Keep the expanders as they were) ...
|
186 |
with st.expander("Learn More: What is a CNN?"):
|
187 |
st.write("""
|
188 |
A Neural Network is a system inspired by the human brain, composed of interconnected nodes (neurons) organized in layers: an input layer, one or more hidden layers, and an output layer.
|
|
|
371 |
st.subheader("Top and Least Performing Classes (by F1-Score)")
|
372 |
with st.container():
|
373 |
top_ten_df = pd.DataFrame(top_ten_dict).sort_values("f1-score", ascending=False)
|
374 |
+
last_ten_df = pd.DataFrame(last_ten_dict).sort_values("f1-score", ascending=True)
|
375 |
|
|
|
376 |
top_ten_df['class_name_display'] = top_ten_df['class_name'].str.replace('_', ' ').str.title()
|
377 |
last_ten_df['class_name_display'] = last_ten_df['class_name'].str.replace('_', ' ').str.title()
|
378 |
|
|
|
379 |
col1, col2 = st.columns(2)
|
380 |
with col1:
|
381 |
st.write("**Top 10 Classes**")
|
382 |
+
st.bar_chart(top_ten_df.set_index('class_name_display')['f1-score'], use_container_width=True)
|
|
|
|
|
383 |
with col2:
|
384 |
st.write("**Bottom 10 Classes**")
|
385 |
+
st.bar_chart(last_ten_df.set_index('class_name_display')['f1-score'], use_container_width=True, color="#ff748c")
|
386 |
+
st.divider() # Divider before the interactive section
|
|
|
|
|
|
|
387 |
|
388 |
# --- Helper Functions ---
|
389 |
+
@st.cache_resource
|
390 |
def load_model(filepath):
|
391 |
"""Loads a Tensorflow Keras Model."""
|
392 |
+
st.write(f"Cache miss: Loading model from {filepath}")
|
393 |
try:
|
394 |
model = tf.keras.models.load_model(filepath)
|
|
|
|
|
395 |
return model
|
396 |
except Exception as e:
|
397 |
st.error(f"Error loading model from {filepath}: {e}")
|
|
|
400 |
def load_prep_image(image_input: UploadedFile, img_shape=224):
|
401 |
"""Reads and preprocesses an image for EfficientNet prediction."""
|
402 |
try:
|
|
|
403 |
bytes_data = image_input.getvalue()
|
|
|
404 |
image_tensor = tf.io.decode_image(bytes_data, channels=3)
|
|
|
|
|
405 |
image_tensor_resized = tf.image.resize(image_tensor, [img_shape, img_shape])
|
|
|
406 |
image_tensor_expanded = tf.expand_dims(image_tensor_resized, axis=0)
|
|
|
|
|
|
|
|
|
|
|
|
|
407 |
return image_tensor_expanded
|
408 |
except Exception as e:
|
409 |
st.error(f"Error processing image: {e}")
|
|
|
426 |
try:
|
427 |
with st.spinner("🤖 Model is predicting..."):
|
428 |
pred_prob = model.predict(processed_image)
|
429 |
+
predicted_index = tf.argmax(pred_prob, axis=1).numpy()[0]
|
430 |
predicted_class_name = class_names[predicted_index]
|
431 |
+
predicted_probability = float(tf.reduce_max(pred_prob).numpy())
|
432 |
return predicted_class_name, predicted_probability
|
433 |
except Exception as e:
|
434 |
st.error(f"Prediction failed: {e}")
|
435 |
return None, None
|
436 |
|
437 |
# --- Interactive Demo Section ---
|
438 |
+
# Header and Caption for the interactive section
|
439 |
st.header(f"Try the Models: :blue[{current_model}] & :blue[{new_model}]")
|
440 |
st.caption("_Model performance may vary. Models are periodically updated._")
|
441 |
|
442 |
+
|
443 |
# Initialize session state keys if they don't exist
|
444 |
if "prediction_result" not in st.session_state:
|
445 |
st.session_state.prediction_result = None
|
|
|
450 |
|
451 |
|
452 |
# Use columns for layout
|
453 |
+
cols = st.columns([3, 0.5, 2, 0.5, 3], gap="medium") # Keep original column ratios
|
454 |
+
|
455 |
|
456 |
# --- Column 1: Image Input ---
|
457 |
with cols[0]:
|
458 |
st.markdown('<div class="centered">', unsafe_allow_html=True) # Apply centering
|
459 |
+
st.subheader("1. Provide an Image") # H3 targeted by CSS
|
460 |
image_source = st.radio(
|
461 |
"Choose image source:",
|
462 |
("Upload Image", "Use Camera"),
|
463 |
key="image_source",
|
464 |
horizontal=True,
|
465 |
+
label_visibility="collapsed"
|
466 |
)
|
467 |
|
468 |
uploaded_image = None
|
|
|
485 |
|
486 |
# Display uploaded image preview
|
487 |
if uploaded_image:
|
488 |
+
image_bytes_for_state = uploaded_image.getvalue()
|
489 |
+
st.image(image_bytes_for_state, caption="Your image", use_column_width='auto')
|
490 |
+
# Removed success message to save space
|
491 |
else:
|
492 |
st.info("Upload or take a picture.")
|
493 |
|
494 |
+
st.markdown('</div>', unsafe_allow_html=True)
|
495 |
+
|
496 |
|
497 |
# --- Column 2: Arrow 1 ---
|
498 |
with cols[1]:
|
499 |
+
# Adjusted padding-top to roughly align arrow with top content
|
500 |
+
st.markdown('<div class="centered" style="justify-content: flex-start; padding-top: 50px;">➡️</div>', unsafe_allow_html=True)
|
501 |
+
|
502 |
|
503 |
# --- Column 3: Model Selection & Prediction ---
|
504 |
with cols[2]:
|
505 |
+
st.markdown('<div class="centered">', unsafe_allow_html=True)
|
506 |
+
st.subheader("2. Select Model") # H3 targeted by CSS
|
507 |
|
508 |
chosen_model = st.radio(
|
509 |
"Pick a Model:",
|
|
|
516 |
model_path_to_use = ""
|
517 |
model_image_path = ""
|
518 |
|
519 |
+
if chosen_model == current_model:
|
520 |
+
model_image_path = "brain.png"
|
521 |
+
model_path_to_use = "model_mini_Food101.keras"
|
522 |
+
elif chosen_model == new_model:
|
523 |
+
model_image_path = "content/creativity_15557951.png"
|
524 |
+
model_path_to_use = "FoodVision.keras"
|
525 |
|
|
|
526 |
try:
|
527 |
if model_image_path:
|
528 |
+
st.image(model_image_path, width=150) # Keep model image
|
529 |
except Exception as e:
|
530 |
st.warning(f"Could not load model image: {model_image_path}")
|
531 |
|
|
|
534 |
label="Predict Food!",
|
535 |
icon="⚛️",
|
536 |
type="primary",
|
537 |
+
use_container_width=True,
|
538 |
+
disabled=not uploaded_image or not model_path_to_use
|
539 |
)
|
540 |
|
541 |
if predict_button:
|
542 |
if uploaded_image and model_path_to_use:
|
|
|
543 |
result_class, result_prob = predict_using_model(uploaded_image, model_path=model_path_to_use)
|
|
|
544 |
st.session_state.prediction_result = result_class
|
545 |
st.session_state.predicted_prob = result_prob
|
546 |
+
st.session_state.predicted_image_bytes = image_bytes_for_state
|
547 |
else:
|
548 |
st.warning("Please provide an image and select a valid model.")
|
549 |
|
550 |
+
st.markdown('</div>', unsafe_allow_html=True)
|
551 |
+
|
552 |
|
553 |
# --- Column 4: Arrow 2 ---
|
554 |
with cols[3]:
|
555 |
+
# Adjusted padding-top
|
556 |
+
st.markdown('<div class="centered" style="justify-content: flex-start; padding-top: 50px;">➡️</div>', unsafe_allow_html=True)
|
557 |
+
|
558 |
|
559 |
# --- Column 5: Output ---
|
560 |
with cols[4]:
|
561 |
+
st.markdown('<div class="centered">', unsafe_allow_html=True)
|
562 |
+
st.subheader("3. Prediction Result") # H3 targeted by CSS
|
563 |
|
|
|
564 |
if st.session_state.prediction_result and st.session_state.predicted_image_bytes:
|
|
|
565 |
st.image(st.session_state.predicted_image_bytes, caption="Image Analyzed", use_column_width='auto')
|
566 |
|
567 |
result_class = st.session_state.prediction_result
|
568 |
probability = st.session_state.predicted_prob
|
569 |
|
|
|
570 |
if "_" in result_class:
|
571 |
modified_class = result_class.replace("_", " ").title()
|
572 |
else:
|
|
|
574 |
|
575 |
st.success(f"Prediction: **:blue[{modified_class}]**")
|
576 |
if probability:
|
577 |
+
st.write(f"Confidence: {probability:.1%}") # Slightly less verbose confidence
|
578 |
|
579 |
elif predict_button:
|
580 |
+
st.error("Prediction failed or image invalid.")
|
|
|
581 |
else:
|
582 |
+
st.info("Result will appear here.")
|
583 |
+
|
584 |
+
st.markdown('</div>', unsafe_allow_html=True)
|
585 |
|
|
|
586 |
|
587 |
# --- Footer or Final Divider ---
|
588 |
+
# st.divider() # Optional: remove if you want less space at the bottom
|