README.md

---
language: en
license: apache-2.0
tags:
- hypervision
- object-detection
- computer-vision
- lightweight
- mobile
- myndlabs
- ncnn
- mnn
- openvino
- android
- pytorch
library_name: hypervision
datasets:
- coco
---

<div align="center">

# HyperVision
**Lightweight anchor-free object detection model. Real-time on mobile devices.**

**Premium Edition by [Myndlabs.tech](https://myndlabs.tech)**

</div>

HyperVision is an anchor-free one-stage object detection model based on Generalized Focal Loss. It is designed for efficient on-device inference across CPU, GPU, and mobile NPU backends.

## Repository Contents

This repository provides:

- **Source code** for training and inference
- **Python demo** for image, video, and webcam inference
- **Android demo** (`demo_android_ncnn/`) using ncnn
- **NCNN C++ demo** (`demo_ncnn/`)
- **MNN C++ demo** (`demo_mnn/`)
- **OpenVINO C++ demo** (`demo_openvino/`)
- **LibTorch C++ demo** (`demo_libtorch/`)
- **Jupyter notebook** walkthrough (`demo/demo-inference-with-pytorch.ipynb`)
- **Multi-backend model export** tools (ONNX, TorchScript)
- **Training pipeline** using PyTorch Lightning

## Install

### Requirements

- Linux, macOS, or Windows
- Python >= 3.7
- PyTorch >= 1.10.0, < 2.0.0

### Quick Start

```shell script
# Clone the repository
git clone https://github.com/Yethikrishna/hypervision.git
cd hypervision

# Install dependencies
pip install -r requirements.txt

# Setup HyperVision
python setup.py develop
```

## Demo

### PyTorch Inference

```bash
# Image inference
python demo/demo.py image --config CONFIG_PATH --model MODEL_PATH --path IMAGE_PATH

# Video inference
python demo/demo.py video --config CONFIG_PATH --model MODEL_PATH --path VIDEO_PATH

# Webcam inference
python demo/demo.py webcam --config CONFIG_PATH --model MODEL_PATH --camid YOUR_CAMERA_ID
```

A Jupyter notebook is also available at `demo/demo-inference-with-pytorch.ipynb`.

### Android

See `demo_android_ncnn/README.md`.

### NCNN, MNN, OpenVINO, LibTorch

See the respective README files in `demo_ncnn/`, `demo_mnn/`, `demo_openvino/`, and `demo_libtorch/`.

## Training

1. Prepare your dataset in COCO, Pascal VOC XML, or YOLO format.
2. Copy and modify a config file from `config/`.
3. Run training:

```shell script
python tools/train.py CONFIG_FILE_PATH
```

TensorBoard logs are saved to the directory specified in the config file.

## Model Export

```shell script
# Export to ONNX
python tools/export_onnx.py --cfg_path CONFIG_PATH --model_path MODEL_PATH

# Export to TorchScript
python tools/export_torchscript.py --cfg_path CONFIG_PATH --model_path MODEL_PATH
```

## Citation

If you use this project in your research, please cite:

```BibTeX
@misc{hypervision,
    title={HyperVision: Lightweight anchor-free object detection model},
    author={Yethikrishna R},
    howpublished = {\url{https://github.com/Yethikrishna/hypervision}},
    year={2025},
    note={Premium edition published by Myndlabs.tech}
}
```

## License

Licensed under the Apache License, Version 2.0. See `LICENSE` for details.

---

**Premium Edition** published by **[Myndlabs.tech](https://myndlabs.tech)** — Enterprise-grade object detection solutions.