Celyrix-1.0-Vision / README.md
mikexsav's picture
Update README.md
9065b7b verified
|
Raw History Blame Contribute Delete
4.33 kB
metadata
license: apache-2.0
license_name: polyform-noncommercial-1.0.0
license_link: LICENSE
datasets:
  - blanchon/EuroSAT_RGB
language:
  - en
pipeline_tag: image-classification
library_name: keras
tags:
  - computer-vision
  - satellite-imagery
  - keras
  - tensorflow
  - remote-sensing
  - land-cover-classification

Celyrix 1.0 Vision

Celyrix 1.0 Vision is a lightweight computer vision model developed by Mirutex Technologies for satellite land-cover classification.

Designed for efficient image analysis, Celyrix identifies visual patterns in RGB satellite imagery and classifies them into five land-cover categories. The model is built around TensorFlow and Keras, with a focus on practical inference, reproducible development, and remote sensing applications.

Celyrix 1.0 Vision represents the first release in the Celyrix model family.

Celyrix Highlights

  • Satellite Vision: Classification of RGB satellite imagery into five land-cover categories.
  • Lightweight Architecture: Designed for efficient inference on conventional computing hardware.
  • Applied Computer Vision: Focused on practical satellite image analysis and geospatial research.
  • Independent Development: Developed by Mirutex Technologies using established machine learning frameworks.
  • Reproducible Research: Built around documented training, evaluation, and deployment workflows.

Model Overview

Property Specification
Model Celyrix 1.0 Vision
Developer Mirutex Technologies
Model Type Image Classification
Architecture Convolutional Neural Network
Framework TensorFlow / Keras
Input RGB satellite imagery
Input Resolution 64 × 64 pixels
Output Five land-cover categories
Dataset EuroSAT RGB
License PolyForm Noncommercial 1.0.0

Classification Categories

The model is designed to recognize five categories derived from EuroSAT:

Category Description
Forest Forest-covered regions
River Rivers and watercourses
Residential Residential buildings and areas
Agriculture Annual crop fields
Industrial Industrial buildings and facilities

The Agriculture category corresponds specifically to the AnnualCrop class in EuroSAT.

Training & Evaluation

Celyrix uses selected classes from the EuroSAT RGB dataset, a satellite imagery classification dataset derived from Sentinel-2 observations.

The development pipeline is based on supervised image classification using TensorFlow and Keras. Model training and evaluation are designed to support reproducible experimentation and consistent preprocessing.

Verified performance metrics, including classification accuracy, precision, recall, F1-score, and confusion matrices, will be documented from actual evaluation results.

Quickstart

Celyrix is distributed through the Hugging Face Hub.

Model repository:

Mirutex/Celyrix-1.0-Vision

Install the core dependencies:

pip install tensorflow keras huggingface_hub pillow numpy

Download the model repository:

from huggingface_hub import snapshot_download

model_dir = snapshot_download(
    repo_id="Mirutex/Celyrix-1.0-Vision"
)

print(model_dir)

Load the published Keras model using the model filename provided in the repository. Apply the documented preprocessing and class mapping before interpreting predictions.

Limitations

Celyrix performs image-level classification and does not provide object detection, geographic localization, or pixel-level segmentation.

The model is designed for satellite imagery similar to its training data. Performance on images from other sensors, geographic distributions, resolutions, or ordinary photography may differ.

Predicted probabilities should not be interpreted as guaranteed confidence estimates.

Celyrix is intended for research, educational experimentation, and applications where outputs can be independently verified.

Acknowledgments

Celyrix uses the EuroSAT RGB dataset, originally introduced by Patrick Helber and collaborators for land-use and land-cover classification using Sentinel-2 imagery.

Reference: EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification