PES-Texture-PCA / app.py
AlirezaF138's picture
Update app.py
f2c754e verified
import numpy as np
import cv2
import gradio as gr
PCA_MODEL_PATH = "pca_texture_model.npy"
COMPONENT_NAMES_PATH = "component_names.txt"
# Load PCA model
pca = np.load(PCA_MODEL_PATH, allow_pickle=True).item()
mean_texture = pca.mean_
components = pca.components_
explained_variance = pca.explained_variance_
n_components = components.shape[0]
TEXTURE_SIZE = int(np.sqrt(mean_texture.shape[0] // 3))
# Calculate slider ranges
slider_ranges = [3 * np.sqrt(var) for var in explained_variance]
# Load component names if available
try:
with open(COMPONENT_NAMES_PATH, "r") as f:
component_names = [f"Component {i+1} ({line.strip()})" if line.strip() else f"Component {i+1}" for i, line in enumerate(f.readlines())]
if len(component_names) < n_components:
component_names += [f"Component {i+1}" for i in range(len(component_names), n_components)]
except FileNotFoundError:
component_names = [f"Component {i+1}" for i in range(n_components)]
def generate_texture(*component_values):
component_values = np.array(component_values)
new_texture = mean_texture + np.dot(component_values, components)
new_texture = np.clip(new_texture, 0, 255).astype(np.uint8)
new_texture = new_texture.reshape((TEXTURE_SIZE, TEXTURE_SIZE, 3))
new_texture = cv2.cvtColor(new_texture, cv2.COLOR_BGR2RGB)
return new_texture
def randomize_texture():
sampled_coefficients = np.random.normal(0, np.sqrt(explained_variance), size=n_components)
return sampled_coefficients.tolist()
def update_texture(*component_values):
texture = generate_texture(*component_values)
return texture
def on_random_click():
random_values = randomize_texture()
texture = generate_texture(*random_values)
updates = [gr.update(value=value) for value in random_values]
updates.append(texture)
return updates
def process_uploaded_image(uploaded_image):
resized_image = cv2.resize(uploaded_image, (TEXTURE_SIZE, TEXTURE_SIZE))
resized_image = cv2.cvtColor(resized_image, cv2.COLOR_RGB2BGR)
flattened_image = resized_image.flatten()
centered_image = flattened_image - mean_texture
coefficients = np.dot(centered_image, components.T)
clipped_coefficients = [np.clip(coeff, -slider_ranges[i], slider_ranges[i]) for i, coeff in enumerate(coefficients)]
return clipped_coefficients
def on_image_upload(image):
coefficients = process_uploaded_image(image)
updates = [gr.update(value=value) for value in coefficients]
return updates
def on_update_click(*component_values):
texture = generate_texture(*component_values)
return texture
# Create Gradio interface
with gr.Blocks() as demo:
with gr.Row():
with gr.Column():
sliders = []
for i in range(n_components):
range_limit = slider_ranges[i]
slider = gr.Slider(
minimum=-range_limit,
maximum=range_limit,
step=10,
value=0,
label=component_names[i]
)
sliders.append(slider)
with gr.Column():
output_image = gr.Image(
label="Generated Texture"
)
upload_image = gr.Image(
label="Upload Image",
sources=['upload', 'clipboard'],
type="numpy"
)
update_texture_button = gr.Button("Update Texture")
random_button = gr.Button("Randomize Texture")
get_components_button = gr.Button("Get Components from Image")
# Update texture when clicking the "Update Texture" button
update_texture_button.click(
fn=on_update_click,
inputs=sliders,
outputs=output_image
)
# Randomize texture and update sliders and image
random_button.click(
fn=on_random_click,
inputs=None,
outputs=[*sliders, output_image]
)
# Update sliders based on the uploaded image when clicking "Get Components from Image"
get_components_button.click(
fn=on_image_upload,
inputs=upload_image,
outputs=sliders
)
# Keep the uploaded image for reference (no update on texture yet)
upload_image.change(
fn=None,
inputs=None,
outputs=[]
)
demo.launch()