---
license: apache-2.0
title: UVIS
sdk: gradio
emoji: 🔥
colorFrom: blue
colorTo: indigo
pinned: true
thumbnail: >-
  https://cdn-uploads.huggingface.co/production/uploads/6820d348853cd8d544c6b014/qapEjDg69wwVgeqCXWTiX.png
short_description: Unified Visual Intelligence System
allow_embedding: true
---

# UVIS - Unified Visual Intelligence System

### A Lightweight Web-Based Visual Perception Demo

> **Try it online**: [uvis.deecoded.io](https://uvis.deecoded.io)
> **GitHub**: [github.com/DurgaDeepakValluri/UVIS](https://github.com/DurgaDeepakValluri/UVIS)

---

## Overview

**UVIS** (Unified Visual Intelligence System) is a **lightweight, web-based visual perception demo**, originally conceptualized as a **spin-off while building Percepta**—a larger modular perception framework.

The goal of UVIS is to make **scene understanding tools more accessible**, allowing anyone to try object detection, semantic segmentation, and depth estimation through a clean web interface, without requiring local setup.

UVIS currently runs on **[Render.com](https://www.render.com)'s Free Tier**, using **lightweight models** to ensure the experience remains stable on limited resources.

---

## Key Features

| Capability                   | Description                                                                         |
| ---------------------------- | ----------------------------------------------------------------------------------- |
| 🟢 **Object Detection**      | YOLOv5-Nano & YOLOv5-Small for fast, low-resource detection.                        |
| 🟢 **Semantic Segmentation** | SegFormer-B0 and DeepLabV3-ResNet50 for general-purpose scenes.                     |
| 🟢 **Depth Estimation**      | MiDaS Small & DPT Lite for per-pixel depth estimation.                              |
| 🖼️ **Scene Blueprint**      | Unified overlay combining all selected tasks.                                       |
| 📊 **Scene Metrics**         | Scene complexity scoring and agent-friendly summaries.                              |
| 📦 **Downloadable Results**  | JSON, overlay images, and ZIP bundles.                                              |
| 🌐 **Web-First Design**      | No installation needed—hosted live at [uvis.deecoded.io](https://uvis.deecoded.io). |
| 🛠️ **Open Source**          | Contribution-friendly, easy to extend and improve.                                  |

---

### Current Limitations & Roadmap

UVIS is designed for **lightweight demos** on **free-tier hosting**, which means:

* Models are optimized for speed and minimal compute.
* Only **image input** is supported at this time.

> As the project grows and higher hosting tiers become available, the roadmap includes:
>
> *  **Video input support**
> *  **Lightweight SLAM**
> *  **Natural language scene descriptions**
> *  **Higher-capacity, more accurate models**

---

## Architecture Highlights

* **Modular Python Backend with Model Registry**
* **Streamlit-Based Interactive Web UI**
* **HuggingFace Transformers & TorchVision Integration**
* **Lightweight Model Support (Render-Compatible)**
* **Structured JSON Output for AI Agents**
* **Robust Error Handling and Logging**

---

## 🤝 Contributing

UVIS is **open-source** and welcomes contributions.
You can:

* Suggest new features
* Improve the web interface
* Extend perception tasks
* Report issues or bugs

### 💻 **Clone and Run Locally**

```bash
git clone https://github.com/DurgaDeepakValluri/UVIS.git
cd UVIS
pip install -r requirements.txt
```

---

## 🌐 Live Demo

> **Explore it online at [uvis.deecoded.io](https://uvis.deecoded.io)**
> Upload an image, select your tasks, and view the results—all in your browser.

---

## 📝 License

Apache 2.0 License. Free for personal and commercial use with attribution.
© 2025 Durga Deepak Valluri