|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- en |
|
|
library_name: transformers |
|
|
tags: |
|
|
- cv |
|
|
- robotics |
|
|
pipeline_tag: object-detection |
|
|
--- |
|
|
# π Celestial-Mini: Lightweight Object Detection Model (TF) |
|
|
|
|
|
[](https://www.tensorflow.org/) |
|
|
[]() |
|
|
[]() |
|
|
|
|
|
**Celestial-Mini** is a compact, high-performance object detection model designed to recognize up to **80 distinct object classes**. Built with **TensorFlow**, it balances speed and accuracy for deployment in edge devices and real-time applications. |
|
|
|
|
|
--- |
|
|
|
|
|
## π Key Features |
|
|
|
|
|
- π Detects up to **80 different object categories** |
|
|
- β‘ Optimized for **real-time inference** |
|
|
- π§ Built on a **lightweight backbone** |
|
|
- π¦ TensorFlow SavedModel format for easy deployment |
|
|
- π§° Compatible with TensorFlow Lite and TensorFlow.js |
|
|
|
|
|
--- |
|
|
|
|
|
## π§ͺ Intended Use |
|
|
|
|
|
Celestial-Mini is designed for: |
|
|
|
|
|
- Robotics and drones |
|
|
- Smart home devices |
|
|
- Augmented Reality (AR) systems |
|
|
- Mobile applications |
|
|
- Educational and prototyping environments |
|
|
|
|
|
--- |
|
|
|
|
|
## π· Object Classes |
|
|
|
|
|
Includes detection support for the standard 80-class COCO-style object categories such as: |
|
|
|
|
|
``` |
|
|
person, bicycle, car, motorcycle, airplane, bus, train, truck, boat, traffic light, ... |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## π¦ How to Use |
|
|
|
|
|
```python |
|
|
import tensorflow as tf |
|
|
|
|
|
# Load the model |
|
|
model = tf.saved_model.load("path/to/celestial-mini") |
|
|
|
|
|
# Run inference |
|
|
detections = model(input_tensor) |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## π Performance |
|
|
|
|
|
| Metric | Value | |
|
|
|----------------|---------------| |
|
|
| Classes | 80 | |
|
|
| Model Size | ~15MB | |
|
|
| Inference Time | < 50ms/image | |
|
|
| Framework | TensorFlow | |
|
|
|
|
|
> π Performance may vary depending on hardware and TensorFlow backend optimizations. |
|
|
|
|
|
--- |
|
|
|
|
|
## π§ Training & Dataset |
|
|
|
|
|
Celestial-Mini was trained on a custom variant of the **COCO dataset**, emphasizing generalization and real-time inference. Model architecture includes quantization-friendly layers and depthwise separable convolutions. |
|
|
|
|
|
--- |
|
|
|
|
|
## π Citation |
|
|
|
|
|
If you use **Celestial-Mini** in your work, please consider citing: |
|
|
|
|
|
``` |
|
|
@misc{celestialmini2025, |
|
|
title={Celestial-Mini: A Lightweight Real-Time Object Detector}, |
|
|
author={Lang, John}, |
|
|
year={2025}, |
|
|
howpublished={\url{https://huggingface.co/langutang/celestial-mini}} |
|
|
} |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## π¬ Contact & License |
|
|
|
|
|
- π« For questions or collaboration, open an issue or contact the maintainer. |
|
|
- βοΈ License: MIT (see LICENSE file for details) |
|
|
|
|
|
--- |
|
|
|
|
|
## π Hugging Face Model Hub |
|
|
|
|
|
To load from Hugging Face: |
|
|
|
|
|
```python |
|
|
from transformers import AutoFeatureExtractor, TFModelForObjectDetection |
|
|
|
|
|
model = TFModelForObjectDetection.from_pretrained("langutang/celestial-mini") |
|
|
extractor = AutoFeatureExtractor.from_pretrained("langutang/celestial-mini") |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
Transform your edge AI projects with the power of **Celestial-Mini** π |