README.md

metadata

license: apache-2.0
metrics:
  - accuracy
pipeline_tag: image-classification
tags:
  - fastai
  - astronomy

Interstellar Classifier

Classify images of interstellar objects such as galaxies, nebulae, comets, asteroids, quasars, and star clusters.

Built for the fast.ai Practical Deep Learning Course by:

• creating a custom dataset (less than 150 images per label) using Bing search API;
• augmenting the dataset; and,
• fine tuning ResNet50 (1 + 3 epochs) in paperspace.com.

Try at: https://huggingface.co/spaces/dcarpintero/fastai-interstellar

Two versions of the model are provided:

Object Model

This version fastai-interstellar-object

Recognizes a (limited) set of specific interstellar objects:

• m31 andromeda
• m33 triangulum
• m81 bode
• m82 cigar
• ngc 1300
• m104 sombrero
• m51 whirlpool
• m42 orion nebula
• m17 omega nebula
• m45 pleiades star cluster

Accuracy: 94.1%

---

Class Model

Available at fastai-interstellar-class

Classifies an image under an astronomy class:

• asteroid
• comet
• galaxy
• nebulae
• planet
• quasar in space
• star cluster

Accuracy: 84%