app.py

# -*- coding: utf-8 -*-
"""FinalProject.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1_wYfP0IRdb9fpc2zvbg8IqdXGx1dTo7X
"""

from datasets import load_dataset
from PIL import Image, ImageChops
from transformers import CLIPProcessor, CLIPModel
from sklearn.metrics.pairwise import cosine_similarity
import torch
import numpy as np
import gradio as gr
from diffusers import StableDiffusionImg2ImgPipeline

# Device setup
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load CLIP model
model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32").to(device)
processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")

# Load dataset
dataset = load_dataset("lirus18/deepfashion", split="train")

# Embed a subset of dataset images
image_vectors = []
image_indices = []
N = 500

for i in range(N):
    img = dataset[i]['image'].convert("RGB")
    inputs = processor(images=img, return_tensors="pt").to(device)
    with torch.no_grad():
        emb = model.get_image_features(**inputs)
    image_vectors.append(emb.cpu().numpy().squeeze())
    image_indices.append(i)

image_vectors = np.array(image_vectors)

# Similarity search
def find_similar(user_image, top_k=3, exclude_index=None):
    inputs = processor(images=user_image.convert("RGB"), return_tensors="pt").to(device)
    with torch.no_grad():
        query_vec = model.get_image_features(**inputs).cpu().numpy()

    sims = cosine_similarity(query_vec, image_vectors)[0]
    if exclude_index is not None:
        sims[exclude_index] = -1

    top_idx = sims.argsort()[-top_k:][::-1]
    return [dataset[image_indices[i]]['image'] for i in top_idx], query_vec

# Load Stable Diffusion
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    low_cpu_mem_usage=True
).to(device)
pipe.enable_attention_slicing()

# Generate 10 images
def generate_outfits(input_image, n=1):
    prompt = "fashion outfit design inspired by the clothing item"
    init_image = input_image.resize((512, 512))
    generated_images = []

    for _ in range(n):
        result = pipe(prompt=prompt, image=init_image, strength=0.7, guidance_scale=7.5)
        generated_images.append(result.images[0])

    return generated_images

# Main function
def recommend_from_upload(uploaded_image):
    uploaded_image = uploaded_image.convert("RGB")

    # Check for duplicates
    closest_idx = None
    for i in range(len(image_indices)):
        dataset_image = dataset[image_indices[i]]['image'].convert("RGB")
        if ImageChops.difference(dataset_image, uploaded_image).getbbox() is None:
            closest_idx = i
            break

    # Find similar items
    similar_imgs, query_vec = find_similar(uploaded_image, top_k=3, exclude_index=closest_idx)

    # Generate 10 new outfits
    generated_imgs = generate_outfits(uploaded_image, n=1)


    # Select best match
    best_score = -1
    best_img = None
    for img in generated_imgs:
        inputs = processor(images=img, return_tensors="pt").to(device)
        with torch.no_grad():
            emb = model.get_image_features(**inputs).cpu().numpy()
        sim = cosine_similarity(query_vec, emb)[0][0]
        if sim > best_score:
            best_score = sim
            best_img = img

    return [uploaded_image] + similar_imgs + [best_img]

# Example paths
example_paths = [
    ["example1.jpg"],
    ["example2.jpg"],
    ["example3.jpg"],
    ["example4.jpg"],
    ["example5.jpg"]
]

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## 👗 Fashion Outfit Recommender")
    gr.Markdown("Upload a clothing image to get 3 similar items from the dataset and 1 AI-generated outfit design.")

    with gr.Row():
        image_input = gr.Image(type="pil", label="Upload a clothing item")

    generate_btn = gr.Button("Generate Recommendations")

    with gr.Row():
        output1 = gr.Image(label="Your Input", height=512, width=384)
        output2 = gr.Image(label="Similar Item 1", height=512, width=384)
        output3 = gr.Image(label="Similar Item 2", height=512, width=384)
        output4 = gr.Image(label="Similar Item 3", height=512, width=384)
        output5 = gr.Image(label="AI-Generated Outfit", height=512, width=384)

    examples = gr.Examples(
        examples=example_paths,
        inputs=image_input,
        label="Try an Example"
    )

    generate_btn.click(fn=recommend_from_upload,
                       inputs=image_input,
                       outputs=[output1, output2, output3, output4, output5])

if __name__ == "__main__":
    demo.launch()