Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import tensorflow as tf
|
3 |
+
import numpy as np
|
4 |
+
from PIL import Image
|
5 |
+
from tensorflow_examples.models.pix2pix import pix2pix
|
6 |
+
|
7 |
+
OUTPUT_CHANNELS = 3
|
8 |
+
|
9 |
+
generator_g = pix2pix.unet_generator(OUTPUT_CHANNELS, norm_type='instancenorm')
|
10 |
+
generator_g.load_weights("vibrantGAN-generator-g-final.weights.h5")
|
11 |
+
|
12 |
+
def preprocess_single_image(image, target_height=256, target_width=256):
|
13 |
+
# Convert PIL image to tensorflow tensor
|
14 |
+
image = tf.convert_to_tensor(np.array(image))
|
15 |
+
|
16 |
+
# Ensure image has 3 channels (RGB)
|
17 |
+
if len(image.shape) == 2: # If grayscale
|
18 |
+
image = tf.stack([image, image, image], axis=-1)
|
19 |
+
elif image.shape[-1] == 4: # If RGBA
|
20 |
+
image = image[:, :, :3]
|
21 |
+
|
22 |
+
# Resize the image
|
23 |
+
image = tf.image.resize(image, [target_height, target_width])
|
24 |
+
|
25 |
+
# Normalize to [-1, 1]
|
26 |
+
image = tf.cast(image, tf.float32)
|
27 |
+
image = (image / 127.5) - 1
|
28 |
+
|
29 |
+
return image
|
30 |
+
|
31 |
+
def process_image(input_image):
|
32 |
+
if input_image is None:
|
33 |
+
return None
|
34 |
+
|
35 |
+
# Get original input image size
|
36 |
+
original_size = input_image.size
|
37 |
+
|
38 |
+
# Preprocess the image
|
39 |
+
processed_input = preprocess_single_image(input_image)
|
40 |
+
|
41 |
+
# Add batch dimension
|
42 |
+
processed_input = tf.expand_dims(processed_input, 0)
|
43 |
+
|
44 |
+
# Generate prediction
|
45 |
+
prediction = generator_g(processed_input)
|
46 |
+
|
47 |
+
# Convert the prediction to displayable format
|
48 |
+
output_image = prediction[0] * 0.5 + 0.5 # Denormalize to [0, 1]
|
49 |
+
output_image = tf.clip_by_value(output_image, 0, 1)
|
50 |
+
|
51 |
+
# Convert to numpy array and then to PIL Image
|
52 |
+
output_array = (output_image.numpy() * 255).astype(np.uint8)
|
53 |
+
output_pil = Image.fromarray(output_array)
|
54 |
+
|
55 |
+
return output_pil
|
56 |
+
|
57 |
+
# Create Gradio interface
|
58 |
+
demo = gr.Interface(
|
59 |
+
fn=process_image,
|
60 |
+
inputs=gr.Image(type="pil", label="Input Image"),
|
61 |
+
outputs=gr.Image(type="pil", label="Generated Output"),
|
62 |
+
title="Image Processing Model",
|
63 |
+
description="Upload an image to see the model's output.",
|
64 |
+
)
|
65 |
+
|
66 |
+
# Launch the interface
|
67 |
+
demo.launch(debug=True)
|