Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import tensorflow as tf
|
3 |
+
from tensorflow.keras.applications import EfficientNetV2L
|
4 |
+
from tensorflow.keras.applications.efficientnet_v2 import preprocess_input, decode_predictions
|
5 |
+
import numpy as np
|
6 |
+
from PIL import Image
|
7 |
+
|
8 |
+
# Lazy loading to optimize memory usage
|
9 |
+
model = None
|
10 |
+
|
11 |
+
def load_model():
|
12 |
+
"""Load the EfficientNetV2L model only when needed."""
|
13 |
+
global model
|
14 |
+
if model is None:
|
15 |
+
model = EfficientNetV2L(weights="imagenet")
|
16 |
+
|
17 |
+
def preprocess_image(image):
|
18 |
+
"""Preprocess the image for EfficientNetV2L model inference."""
|
19 |
+
image = image.resize((480, 480)) # Resize for EfficientNetV2L
|
20 |
+
image_array = np.array(image) # Convert to NumPy array
|
21 |
+
image_array = preprocess_input(image_array) # Normalize input
|
22 |
+
image_array = np.expand_dims(image_array, axis=0) # Add batch dimension
|
23 |
+
return image_array
|
24 |
+
|
25 |
+
def predict_image(image):
|
26 |
+
"""
|
27 |
+
Process the uploaded image and return the top 3 predictions.
|
28 |
+
"""
|
29 |
+
try:
|
30 |
+
load_model() # Ensure the model is loaded
|
31 |
+
image_array = preprocess_image(image) # Preprocess image
|
32 |
+
predictions = model.predict(image_array) # Get predictions
|
33 |
+
decoded_predictions = decode_predictions(predictions, top=3)[0]
|
34 |
+
|
35 |
+
# Format predictions as a dictionary (label -> confidence)
|
36 |
+
return {label: float(confidence) for _, label, confidence in decoded_predictions}
|
37 |
+
|
38 |
+
except Exception as e:
|
39 |
+
return {"Error": str(e)}
|
40 |
+
|
41 |
+
# Create the Gradio interface
|
42 |
+
interface = gr.Interface(
|
43 |
+
fn=predict_image,
|
44 |
+
inputs=gr.Image(type="pil"), # Accepts an image input
|
45 |
+
outputs=gr.Label(num_top_classes=3), # Shows top 3 predictions
|
46 |
+
title="EfficientNetV2L Image Classifier",
|
47 |
+
description="Upload an image, and the model will predict its content with high accuracy.",
|
48 |
+
allow_flagging="never" # Disable flagging to avoid unnecessary logs
|
49 |
+
)
|
50 |
+
|
51 |
+
# Launch the Gradio app
|
52 |
+
if __name__ == "__main__":
|
53 |
+
interface.launch()
|