Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from PIL import Image
|
3 |
+
import torch
|
4 |
+
from transformers import pipeline
|
5 |
+
from torchvision import models, transforms
|
6 |
+
|
7 |
+
# Load the models
|
8 |
+
text_to_image_pipeline = pipeline("text-to-image-generation", model="CompVis/stable-diffusion-v1-4")
|
9 |
+
segmentation_model = models.segmentation.deeplabv3_resnet101(pretrained=True)
|
10 |
+
segmentation_model.eval()
|
11 |
+
|
12 |
+
# Define transformation for the segmentation model
|
13 |
+
preprocess = transforms.Compose([
|
14 |
+
transforms.Resize(256),
|
15 |
+
transforms.CenterCrop(224),
|
16 |
+
transforms.ToTensor(),
|
17 |
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
18 |
+
])
|
19 |
+
|
20 |
+
# Helper function to segment clothing area
|
21 |
+
def segment_clothing(image):
|
22 |
+
input_tensor = preprocess(image)
|
23 |
+
input_batch = input_tensor.unsqueeze(0)
|
24 |
+
|
25 |
+
with torch.no_grad():
|
26 |
+
output = segmentation_model(input_batch)['out'][0]
|
27 |
+
output_predictions = output.argmax(0)
|
28 |
+
mask = output_predictions.byte().cpu().numpy()
|
29 |
+
|
30 |
+
return mask
|
31 |
+
|
32 |
+
# Function to generate base image
|
33 |
+
def generate_base_image(base_prompt_part1, base_prompt_color, base_prompt_clothing):
|
34 |
+
# Combine the parts to create the full base prompt
|
35 |
+
base_prompt = f"{base_prompt_part1} {base_prompt_color} {base_prompt_clothing}"
|
36 |
+
|
37 |
+
# Generate base clothing image
|
38 |
+
base_image = text_to_image_pipeline(base_prompt)[0]
|
39 |
+
base_image = Image.fromarray(base_image)
|
40 |
+
|
41 |
+
return base_image
|
42 |
+
|
43 |
+
# Define the function to generate design and paste it on the clothing
|
44 |
+
def generate_and_paste_design(base_image, design_prompt):
|
45 |
+
# Generate design
|
46 |
+
generated_image = text_to_image_pipeline(design_prompt)[0]
|
47 |
+
generated_design = Image.fromarray(generated_image)
|
48 |
+
|
49 |
+
# Segment the clothing area
|
50 |
+
clothing_mask = segment_clothing(base_image)
|
51 |
+
|
52 |
+
# Ensure the generated design fits within the clothing area
|
53 |
+
generated_design = generated_design.resize(base_image.size)
|
54 |
+
|
55 |
+
# Paste the design onto the clothing area
|
56 |
+
clothing_area = Image.composite(generated_design, base_image, Image.fromarray(clothing_mask*255))
|
57 |
+
|
58 |
+
return clothing_area
|
59 |
+
|
60 |
+
# Create the Gradio interface
|
61 |
+
base_prompt_part1_input = gr.inputs.Textbox(lines=1, placeholder="Enter 'a single plain'")
|
62 |
+
base_prompt_color_input = gr.inputs.Textbox(lines=1, placeholder="Enter color type")
|
63 |
+
base_prompt_clothing_input = gr.inputs.Textbox(lines=1, placeholder="Enter clothing type")
|
64 |
+
design_prompt_input = gr.inputs.Textbox(lines=1, placeholder="Enter design prompt")
|
65 |
+
output_image = gr.outputs.Image(type="pil")
|
66 |
+
|
67 |
+
def full_process(base_prompt_part1, base_prompt_color, base_prompt_clothing, design_prompt):
|
68 |
+
# Generate the base image
|
69 |
+
base_image = generate_base_image(base_prompt_part1, base_prompt_color, base_prompt_clothing)
|
70 |
+
|
71 |
+
# Generate and paste the design on the base image
|
72 |
+
final_image = generate_and_paste_design(base_image, design_prompt)
|
73 |
+
|
74 |
+
return final_image
|
75 |
+
|
76 |
+
gr.Interface(
|
77 |
+
fn=full_process,
|
78 |
+
inputs=[base_prompt_part1_input, base_prompt_color_input, base_prompt_clothing_input, design_prompt_input],
|
79 |
+
outputs=output_image,
|
80 |
+
title="Design and Paste on Clothing",
|
81 |
+
description="Generate a base clothing image from the given prompts and paste the generated design onto it."
|
82 |
+
).launch()
|