app.py

import gradio as gr
import os
from torchvision.transforms import Resize
from upstash_vector import Index
from datasets import load_dataset
from transformers import AutoFeatureExtractor, AutoModel

index = Index.from_env()
print(os.environ["UPSTASH_VECTOR_REST_URL"])
print(os.environ["UPSTASH_VECTOR_REST_TOKEN"])

resize_transform = Resize((250, 250))

model_ckpt = "google/vit-base-patch16-224-in21k"
extractor = AutoFeatureExtractor.from_pretrained(model_ckpt)
model = AutoModel.from_pretrained(model_ckpt)
hidden_dim = model.config.hidden_size

dataset = load_dataset("HengJi/human_faces")

with gr.Blocks() as demo:
    gr.Markdown(
        """
        # Find Your Twins

        Upload your face and find the most similar people from the X dataset. Powered by [Upstash Vector](https://upstash.com) 🚀
        """
    )

    with gr.Tab("Basic"):
        with gr.Row():
            with gr.Column(scale=1):
                input_image = gr.Image(type="pil")
            with gr.Column(scale=3):
                output_image = gr.Gallery()


        @input_image.upload(inputs=input_image, outputs=output_image)
        def find_similar_faces(image):
            resized_image = resize_transform(image)
            inputs = extractor(images=resized_image, return_tensors="pt")
            outputs = model(**inputs)
            embed = outputs.last_hidden_state[0][0]
            result = index.query(vector=embed.tolist(), top_k=3)
            return [dataset["train"][int(vector.id[3:])]["image"] for vector in result]

    with gr.Tab("Advanced"):
        with gr.Row():
            with gr.Column(scale=1):
                adv_input_image = gr.Image(type="pil")
                adv_image_count = gr.Number(9, label="Image Count")

            with gr.Column(scale=3):
                adv_output_image = gr.Gallery(height=1000)


        @adv_input_image.upload(inputs=[adv_input_image, adv_image_count], outputs=[adv_output_image])
        def find_similar_faces(image, count):
            resized_image = resize_transform(image)
            inputs = extractor(images=resized_image, return_tensors="pt")
            outputs = model(**inputs)
            embed = outputs.last_hidden_state[0][0]
            result = index.query(vector=embed.tolist(), top_k=min(count, 9))
            return [dataset["train"][int(vector.id[3:])]["image"] for vector in result]

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