1. The IncliNET Tutorial

1.1. Introduction

An application with and online web GUI and an easy to access API for determining the inclinations of the spiral galaxies using their optical images.

This program is powered by several deep convolutional neural networks (CNN) implemented in TensorFlow. All CNNs are constructed based on the well-known VGG structure, where the convolutional filters are of size 3x3.

1.1.1. Problem

The inclination of spiral galaxies plays an important role in measurements of their distances using the Tully-Fisher relationship. Each galaxy has its own unique morphology, luminosity, and surface brightness profiles. In addition, galaxy images are covered by foreground stars of the Milky Way galaxy. Therefore, it is challenging to design an algorithm that automatically determines the 3D inclination of spiral galaxies. The inclinations of spiral galaxies can be coarsely derived from the ellipticity of apertures used for photometry, assuming that the image of a spiral galaxy is the projection of a disk with the shape of an oblate spheroid. For ~1/3 of spirals, the approximation of axial ratios provides inclination estimates good to better than 5 degrees, with degradation to ~5 degrees for another 1/3. However in ~1/3 of cases, ellipticity-deriven inclinations are problematic for a variety of reasons. Prominent bulges can dominate the axial ratio measurement, with the Sombrero galaxy (scroll down to see an image of The Sombrero Galaxy) providing an extreme example. Some galaxies may not be axially symmetric due to tidal effects. High surface brightness bars within much lower surface brightness disks can lead to large errors. Simply the orientation of strong spiral features with respect to the tilt axis can be confusing. The statistical derivation of inclinations for large samples has been unsatisfactory.

1.1.2. Objectives

The task of manual evaluation of spirals inclinations is tedious and time consuming. In the future, with the large astronomical survey telescopes coming online, this task could not be efficiently done manually for thousands of galaxies. The objective of this project is to automatically determine the inclination of spiral galaxies with the human level accuracy providing their images (ideally in both colorful and black-and-white formats).

2. Web Application

https://edd.ifa.hawaii.edu/inclinet/

On the left side of this tool, users have different options to find and load a galaxy image. The PGC-based query relies on the information provided by the HyperLEDA catalog. Each image is rotated and resized based on the LEDA entries for logd25 and position angle (pa), which are reasonable in most cases. Further manual alignment features are provided, however the evaluation process is independent of the orientation of the image. Clicking on the Evaluate button, the output inclinations generated by various ML models are generated and the average results are displayed on the right side. This step feeds the image to a pre-trained neural network(s) and outputs the averages of the determined inclination value. In addition, there is other networks that separately predict the rejection probability of the galaxy by human users. For practical reasons, all images are converted to square sizes and rescaled to 128x128 pixels prior to the evaluation process.

This online GUI allows users to submit a galaxy image through four different methods, as described below.

2.1. Galaxy PGC ID

Entering the name of a galaxy by querying its PGC number (the ID of galaxy in the Principal Galaxy Catalog) - The PGC catalog is deployed with our model, and contains a table of galaxy coordinates and their sizes. Images are then queried from the SDSS quick-look image server.

2.2. Galaxy Name

Searching a galaxy by its common name. - The entered name is queried through the NASA/IPAC Extragalactic Database. Then, a python routine based on the package Beautiful Soup extracts the corresponding PGC number. Once the PGC ID is available, the galaxy image is imported from the SDSS quick-look as explained above.

2.3. Galaxy Coordinates

Looking at a specific location in the sky by entering the sky coordinates and the field size. In the first release we only provide access to the SDSS images, if they are available. The SDSS coverage is mainly limited to the Northern sky.

2.4. Galaxy Image

Uploading a galaxy image from the local computer of the user. - User has the option of uploading a galaxy image for evaluation by our model(s)

3. API

We provide a REST API that can be called from terminal or other applications that enable the REST API calls. The input and output data would be in JSON format.

3.1. Example 1: /api/pgc/

3.1.1. Sending a request with the galaxy ID

Evaluating inclination based on the PGC ID of the target galaxy. Here we test the API for pgc=69327.

• On average, there is ~7% that a human user rejects this galaxy.

• The median of all evaluated inclinations is 75 degree (from face-on).

import matplotlib.pyplot as plt
from PIL import Image

img = Image.open('./PGC69327_NGC7331.png')
plt.imshow(img)
plt.title("NGC7331")

Text(0.5, 1.0, 'NGC7331')

! curl http://edd.ifa.hawaii.edu/inclinet/api/pgc/69327

{
  "status": "success",
  "galaxy": {
    "pgc": "69327",
    "ra": "339.2673 deg",
    "dec": "34.4156 deg",
    "fov": "14.00 arcmin",
    "pa": "169.0 deg",
    "objname": "NGC7331"
  },
  "inclinations": {
    "Group_0": {
      "model4": 75.0,
      "model41": 76.0,
      "model42": 74.0,
      "model43": 74.0
    },
    "Group_1": {
      "model5": 76.0,
      "model51": 76.0,
      "model52": 75.0,
      "model53": 73.0
    },
    "Group_2": {
      "model6": 75.0,
      "model61": 77.0,
      "model62": 76.0,
      "model63": 75.0
    },
    "summary": {
      "mean": 75.16666666666667,
      "median": 75.0,
      "stdev": 1.0671873729054746
    }
  },
  "rejection_likelihood": {
    "model4-binary": 13.275855779647827,
    "model5-binary": 0.4835432395339012,
    "model6-binary": 2.4772919714450836,
    "summary": {
      "mean": 5.412230330208938,
      "median": 2.4772919714450836,
      "stdev": 5.619680442177983
    }
  }
}

3.1.2. Erroneous PGC ID

or an ID that doesn’t exist

! curl http://edd.ifa.hawaii.edu/inclinet/api/pgc/xxxx

{"message":"Could not find PGCxxxx in the database. ","status":"failed"}

3.2. Example 2: /api/objname

3.2.1. Sending a request with the galaxy name

Evaluating inclination based on the Galaxy Name. Here we test the API for ‘Sombrero’.

• On average, there is ~44% that a human user rejects this galaxy. However the descrepancies between different models is quites large, as 1\(\sigma\) standard devaition of the rejection likelihoods is ~20%. It would be suggested that a human overlooks the automatically determined inclination for this galaxy.

• Click here for the manual evaluations on Galaxy Inclination Zoo.

• The median of all evaluated inclinations is 89 degree (from face-on).

import matplotlib.pyplot as plt
from PIL import Image

img = Image.open('./sombrero.png')
plt.imshow(img)
plt.title("The Sombrero Galaxy (M104)")

Text(0.5, 1.0, 'The Sombrero Galaxy (M104)')

! curl http://edd.ifa.hawaii.edu/inclinet/api/objname/Sombrero

{
  "status": "success",
  "galaxy": {
    "pgc": 42407,
    "ra": "189.9977 deg",
    "dec": "-11.6230 deg",
    "fov": "12.77 arcmin",
    "pa": "89.5 deg",
    "objname": "Sombrero"
  },
  "inclinations": {
    "Group_0": {
      "model4": 89.0,
      "model41": 89.0,
      "model42": 75.0,
      "model43": 85.0
    },
    "Group_1": {
      "model5": 87.0,
      "model51": 90.0,
      "model52": 88.0,
      "model53": 88.0
    },
    "Group_2": {
      "model6": 90.0,
      "model61": 90.0,
      "model62": 88.0,
      "model63": 90.0
    },
    "summary": {
      "mean": 87.41666666666667,
      "median": 88.5,
      "stdev": 4.009537241239803
    }
  },
  "rejection_likelihood": {
    "model4-binary": 59.36983823776245,
    "model5-binary": 14.682640135288239,
    "model6-binary": 60.917991399765015,
    "summary": {
      "mean": 44.99015659093857,
      "median": 59.36983823776245,
      "stdev": 21.439968280999413
    }
  }
}

3.2.2. Wrong galaxy name

or a name that can not be recognized/parsed from NED.

! curl http://edd.ifa.hawaii.edu/inclinet/api/objname/myGalaxy

{"message":"Could not find oject myGalaxy.","status":"failed"}

3.3. Example 3: /api/file

3.3.1. Sending a request with the galaxy image

Evaluating inclination using the Galaxy Image. Here, we test the API for ‘Andromeda’.

• On average, there is ~4% that a human user rejects this galaxy.

• The median of all evaluated inclinations is 76 degree (from face-on).

import matplotlib.pyplot as plt
from PIL import Image

img = Image.open('./andromeda.jpg')
plt.imshow(img)
plt.title("Andromeda")

Text(0.5, 1.0, 'Andromeda')

! curl -F 'file=@./andromeda.jpg' http://edd.ifa.hawaii.edu/inclinet/api/file

{
  "status": "success",
  "filename": "andromeda.jpg",
  "inclinations": {
    "Group_0": {
      "model4": 77.0,
      "model41": 79.0,
      "model42": 76.0,
      "model43": 76.0
    },
    "Group_1": {
      "model5": 73.0,
      "model51": 76.0,
      "model52": 75.0,
      "model53": 72.0
    },
    "Group_2": {
      "model6": 78.0,
      "model61": 76.0,
      "model62": 77.0,
      "model63": 78.0
    },
    "summary": {
      "mean": 76.08333333333333,
      "median": 76.0,
      "stdev": 1.9346977943739831
    }
  },
  "rejection_likelihood": {
    "model4-binary": 9.210722893476486,
    "model5-binary": 34.75632667541504,
    "model6-binary": 8.197003602981567,
    "summary": {
      "mean": 17.3880177239577,
      "median": 9.210722893476486,
      "stdev": 12.28821993007762
    }
  }
}

3.3.2. Wrong file name

! curl -F 'file=@./myGalaxy.jpg' http://edd.ifa.hawaii.edu/inclinet/api/file

curl: (26) Failed to open/read local data from file/application

3.3.3. Wrong file format

! curl -F 'file=@./myGalaxy.txt' http://edd.ifa.hawaii.edu/inclinet/api/file

{"filename":"myGalaxy.txt","message":"[Note] Please upload jpg, png, jpeg, and gif giles no larger than 1 MB !","status":"failed"}

3.4. Example 4: /getPGC

3.4.1. Get information about the galaxy providing the PGC ID

PGC 2557 is the code of the Andromeda galaxy.

! curl -X 'POST' \
      'http://edd.ifa.hawaii.edu/inclinet/getPGC' \
      -H 'accept: application/json' \
      -H 'Content-Type: application/json' \
      -d '{"pgc": 2557}'

{"dec":"41.2689","fov":"266.74","objname":"NGC0224","pa":"35.0","pgc":2557,"ra":"10.6848","status":"success"}

3.4.2. Generating a wrong API call

Cases when the requested PGC ID is not available, or an error occurs in the API call

! curl -X 'POST' \
      'http://edd.ifa.hawaii.edu/inclinet/getPGC' \
      -H 'accept: application/json' \
      -H 'Content-Type: application/json' \
      -d '{"pgc": 20505070}'

{"status":"error"}

3.5. Example 5: /getObj

3.5.1. Get information about the galaxy providing its common name

M31 is the alternative name of Andromeda in the Messier catalog.

! curl -X 'POST' \
      'http://edd.ifa.hawaii.edu/inclinet/getObj' \
      -H 'accept: application/json' \
      -H 'Content-Type: application/json' \
      -d '{"objname": "M31"}'

{"dec":"41.2689","fov":"266.74","objname":"M31","pa":"35.0","pgc":2557,"ra":"10.6848","status":"success"}

3.5.2. Generating a wrong API call

Cases when the requested galaxy name is not available, or an error occurs in the API call

! curl -X 'POST' \
      'http://edd.ifa.hawaii.edu/inclinet/getObj' \
      -H 'accept: application/json' \
      -H 'Content-Type: application/json' \
      -d '{"objname": "myGalaxy"}'

{"status":"error"}

3.6. Related Topics

• For further details on various VGG models we considered in this project click here.

• Visit here to get the full picture of the deployment plan.

• The Production Pipeline

• Project proposal and motivations

• Data Preprocessing in order to get reliable labels

• On how to download data from the SDSS image service and preprocess them click here

4. Acknowledgments

4.1. About the data

All data exposed by the IncliNET project belongs to

• Cosmicflows-4 program

• Copyright (C) Cosmicflows

• Team - The Extragalactic Distance Database (EDD)

4.2. Citation

Please cite the following paper and the gitHub repository of this project.

Cosmicflows-4: The Catalog of ∼10,000 Tully-Fisher Distances (2020, ApJ, 902, 145)

@ARTICLE{2020ApJ...902..145K,
       author = {{Kourkchi}, Ehsan and {Tully}, R. Brent and {Eftekharzadeh}, Sarah and {Llop}, Jordan and {Courtois}, H{\'e}l{\`e}ne M. and {Guinet}, Daniel and {Dupuy}, Alexandra and {Neill}, James D. and {Seibert}, Mark and {Andrews}, Michael and {Chuang}, Juana and {Danesh}, Arash and {Gonzalez}, Randy and {Holthaus}, Alexandria and {Mokelke}, Amber and {Schoen}, Devin and {Urasaki}, Chase},
        title = "{Cosmicflows-4: The Catalog of {\ensuremath{\sim}}10,000 Tully-Fisher Distances}",
      journal = {\apj},
     keywords = {Galaxy distances, Spiral galaxies, Galaxy photometry, Hubble constant, H I line emission, Large-scale structure of the universe, Inclination, Sky surveys, Catalogs, Distance measure, Random Forests, 590, 1560, 611, 758, 690, 902, 780, 1464, 205, 395, 1935, Astrophysics - Astrophysics of Galaxies},
         year = 2020,
        month = oct,
       volume = {902},
       number = {2},
          eid = {145},
        pages = {145},
          doi = {10.3847/1538-4357/abb66b},
archivePrefix = {arXiv},
       eprint = {2009.00733},
 primaryClass = {astro-ph.GA},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2020ApJ...902..145K},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}

4.3. Author

• Ehsan Kourkchi - ekourkchi@gmail.com

4.4. Disclaimer

• All rights reserved. The material may not be used, reproduced or distributed, in whole or in part, without the prior agreement.