Model Card for Wu, cloud identifier

The model identifies clouds from ground-based images of the sky.

Model Details

Model Description

This is a simple feedforward neural network for image classification.

The model was trained on an open dataset provided by Liu, Pu (2019), first used by Zhang et. al (2018)

Uses

The model is intended to classify ground-based photos of the sky containing clouds. These photos should be taken during the day, with the sun providing sufficient lighting.

Out-of-Scope Use

The model will not work well with satellite-based imagery. The model will probably not work well with nighttime photos as the lighting is sufficiently distinct.

How to Get Started with the Model

TODO

Training Details

Training Data

See https://huggingface.co/datasets/aduuuuuu/CCSN. This is a collection of 2,543 ground-based images of the sky. Photos are 400x400 although some are 256x256. These smaller images are upscaled. All images are converted to greyscale.

Evaluation

Dataset	Loss	Accuracy	Runtime	Steps
Train	2.5057	0.2285	13s	32
Test	2.4874	0.2441	13s	8