app.py

import gradio as gr
import pickle, os
import pandas as pd
import numpy as np
import os
from transformers import CLIPProcessor, CLIPModel
from datasets import load_dataset

from PIL import Image
import requests
from io import BytesIO

model = CLIPModel.from_pretrained("patrickjohncyh/fashion-clip")
processor = CLIPProcessor.from_pretrained("patrickjohncyh/fashion-clip")

# hf_token = os.environ.get("HF_API_TOKEN")
# dataset = load_dataset('pratyush19/cyborg', use_auth_token=hf_token, split='train')
 
# dir_path = "train/"
# print (dataset)
# print (dataset[0].keys())

with open('valid_images_sample.pkl','rb') as f:
    valid_images = pickle.load(f)

with open('image_encodings_sample.pkl','rb') as f:
    image_encodings = pickle.load(f)
valid_images = np.array(valid_images)

with open('PIL_images.pkl','rb') as f:
    PIL_images = pickle.load(f)


def softmax(x):
    e_x = np.exp(x - np.max(x, axis=1, keepdims=True))
    return e_x / e_x.sum(axis=1, keepdims=True)

def find_similar_images(caption, image_encodings):
    inputs = processor(text=[caption], return_tensors="pt")
    text_features = model.get_text_features(**inputs)
    text_features = text_features.detach().numpy()
    logits_per_image = softmax(np.dot(text_features, image_encodings.T))
    return logits_per_image

def find_relevant_images(caption):
    similarity_scores = find_similar_images(caption, image_encodings)[0]
    top_indices = np.argsort(similarity_scores)[::-1][:16]
#     top_path = valid_images[top_indices]
    images = []
    for idx in top_indices:
        images.append(PIL_images[idx])
    return images

def gradio_interface(input_text):
    # with open("user_inputs.txt", "a") as file:
    #     file.write(input_text + "\n")
    images = find_relevant_images(input_text)
    return images


def clear_inputs():
    return [None, None, None, None, None, None, None, None, None,
            None, None, None, None, None, None, None, None, None]

outputs = [None]*16

with gr.Blocks(title="MirrAI") as demo:
    gr.Markdown("<h1 style='text-align: center;'>MirrAI: GenAI-based Fashion Search</h1>")
    gr.Markdown("Enter a text to find the most relevant images from our dataset.")
        
    text_input = gr.Textbox(lines=1, label="Input Text", placeholder="Enter your text here...")
    with gr.Row():
        cancel_button = gr.Button("Cancel")
        submit_button = gr.Button("Submit")
    examples = gr.Examples(["high-rise flare jean",
                            "a-line dress with floral",
                            "men colorful blazers",
                            "jumpsuit with puffed sleeve",
                            "sleeveless sweater",
                            "floral shirt",
                            "blue asymmetrical wedding dress with one sleeve",
                            "women long coat",
                            "cardigan sweater"], inputs=[text_input])

    with gr.Row():
        outputs[0] = gr.Image()
        outputs[1] = gr.Image()
        outputs[2] = gr.Image()
        outputs[3] = gr.Image()
    with gr.Row():
        outputs[4] = gr.Image()
        outputs[5] = gr.Image()
        outputs[6] = gr.Image()
        outputs[7] = gr.Image()
    with gr.Row():
        outputs[8] = gr.Image()
        outputs[9] = gr.Image()
        outputs[10] = gr.Image()
        outputs[11] = gr.Image()
    with gr.Row():
        outputs[12] = gr.Image()
        outputs[13] = gr.Image()
        outputs[14] = gr.Image()
        outputs[15] = gr.Image()
    
    submit_button.click(
        fn=gradio_interface,
        inputs=text_input,
        outputs=outputs
    )
    cancel_button.click(
        fn=clear_inputs,
        inputs=None,
        outputs=[text_input] + outputs
    )

demo.launch(share=True)