initial commit
Browse files- app.py +57 -0
- autoencoder.h5 +3 -0
- decoder.h5 +3 -0
- requirements.txt +4 -0
app.py
ADDED
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import numpy as np
|
2 |
+
import gradio as gr
|
3 |
+
from tensorflow.keras.models import load_model
|
4 |
+
|
5 |
+
# Carga los modelos previamente entrenados
|
6 |
+
autoencoder = load_model("autoencoder.h5")
|
7 |
+
decoder = load_model("decoder.h5")
|
8 |
+
|
9 |
+
latent_dim = 128
|
10 |
+
|
11 |
+
def add_gaussian_noise(image, noise_factor=0.2):
|
12 |
+
noisy_image = image + noise_factor * np.random.normal(size=image.shape)
|
13 |
+
noisy_image = np.clip(noisy_image, 0., 1.)
|
14 |
+
return noisy_image
|
15 |
+
|
16 |
+
def denoise_and_generate(image, num_images):
|
17 |
+
image = np.array(image) / 255.0
|
18 |
+
noisy_image = add_gaussian_noise(np.expand_dims(image, axis=0))
|
19 |
+
denoised_image = autoencoder.predict(noisy_image).squeeze()
|
20 |
+
denoised_image = (denoised_image * 255).astype(np.uint8)
|
21 |
+
|
22 |
+
# Genera im谩genes con el VAE
|
23 |
+
random_latent_vectors = np.random.normal(size=(num_images, latent_dim))
|
24 |
+
generated_images = decoder.predict(random_latent_vectors)
|
25 |
+
|
26 |
+
# Prepara las im谩genes para devolverlas
|
27 |
+
outputs = [denoised_image] + [generated_images[i].squeeze() for i in range(num_images)]
|
28 |
+
|
29 |
+
# Rellena los valores faltantes para que siempre haya 11 valores
|
30 |
+
outputs += [None] * (11 - len(outputs))
|
31 |
+
|
32 |
+
return outputs
|
33 |
+
|
34 |
+
# Define la interfaz
|
35 |
+
inputs = [
|
36 |
+
gr.Image(label="Imagen de Entrada"),
|
37 |
+
gr.Slider(1, 2, step=1, label="N煤mero de Im谩genes Generadas")
|
38 |
+
]
|
39 |
+
|
40 |
+
outputs = [
|
41 |
+
gr.Image(label="Imagen Reconstruida (DAE)"),
|
42 |
+
] + [
|
43 |
+
gr.Image(label=f"Imagen Generada {i+1} (VAE)") for i in range(2)
|
44 |
+
]
|
45 |
+
|
46 |
+
# Crea la aplicaci贸n Gradio
|
47 |
+
interface = gr.Interface(
|
48 |
+
fn=denoise_and_generate,
|
49 |
+
inputs=inputs,
|
50 |
+
outputs=outputs,
|
51 |
+
title="Interfaz Interactiva para DAE y VAE",
|
52 |
+
description="Sube una imagen para denoising con DAE y genera im谩genes nuevas con VAE."
|
53 |
+
)
|
54 |
+
|
55 |
+
# Lanza la aplicaci贸n
|
56 |
+
if __name__ == "__main__":
|
57 |
+
interface.launch(share=True)
|
autoencoder.h5
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:9d917367556b4ece04e36c71ef36a9e341e397c22a60f108a24e01835db3bf64
|
3 |
+
size 991848
|
decoder.h5
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:904e96c97e027845f268e66ef0b43ab40d37b624ce1fdfe581e5782132531742
|
3 |
+
size 27492272
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
tensorflow>=2.0.0
|
2 |
+
gradio>=3.50.0
|
3 |
+
numpy>=1.19.0
|
4 |
+
pillow>=9.0.0
|