README.md

---
title: Space
emoji: 🏃
colorFrom: red
colorTo: yellow
sdk: gradio
sdk_version: 5.48.0
app_file: app.py
pinned: false
short_description: James Webb
thumbnail: >-
  https://cdn-uploads.huggingface.co/production/uploads/68dc8c5a7c207d5db359cbb9/foCePE-ZvhB3ozg1wqZ-C.webp
---

## Overview
This app demonstrates a **multimodal AI search tool** using both **natural language processing** and **computer vision**.  
It allows users to search an index of 1,000 images using either a text query, an image upload, or both.  
The model used (CLIP) embeds text and images in a shared vector space so that semantic similarity can be compared directly.

## How to Use
1. Wait for the “Index built: 1000 images” message.  
2. Enter a **text query** (e.g., “spiral galaxy”) or upload an **image**.  
3. Adjust the **Top K slider** to set how many top matches to view.  
4. Click **Search** to see the results ranked by similarity score.  
5. The grid displays the most relevant images first.

## About the Model
- **Model:** CLIP (Contrastive Language–Image Pre-training)  
- **Capabilities:** Combines natural-language understanding with visual feature recognition.  
- **Purpose:** Demonstrates integration of NLP and computer vision in a single multimodal application.  

## Evaluation Summary
A brief qualitative test on 10 queries showed that roughly **85 % of the top-5 results** were visually relevant.  
This confirms that the embeddings correctly align text and image meanings.

## Limitations
- Works best with visually distinctive subjects (e.g., planets, galaxies).  
- No fine-tuning on this dataset.  
- Index must be rebuilt if files are changed unless persistence is added.

## Credits
- **Dataset:** NASA James Webb Telescope image collection  
- **Model Source:** [Hugging Face CLIP](https://huggingface.co/openai/clip-vit-base-patch32)  
- **Created by:** Jay McIntyre for UMGC ARIN-460 Assignment 8  

---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference