Awards & Nominations
Exo-Vision has received the following awards and nominations. Way to go!
Global Nominee
ExoVision
High-Level Project Summary
Using NASA exoplanets archive, and using different known parameters such as radius, mass, star type etc. from different planetary systems we generated an image of the landscape of the planet as it's believed to look like (as far as current scientific knowledge tells us) using 'stable diffusion' ,an open source project trained by us , and properties generator code, written in python, that gives the planet properties.The user interface is fully on the web. The user has the option to enter a specific name of a known planet name from the data base. For users who are not familiar with planet name there's auto complete implemented, or defining his own planet. The site has extra explanations.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
To make the Image generator you have just used we needed two main things, the first and most important one is finding the connection between the different exo-planets parameters (radius, mass, star type, etc…) to his expected/believed visual representation.
The second thing was to have a way to make an AI generated image quickly. To do so we used an open source API (Application programming interface), called ‘Stable Diffusion’ ,which is an open source code that receives a text, and using AI (Artificial intelligence) tries to give back an image. This technology is quite new and it doesn’t work perfectly yet, meaning a lot of times the AI returns inaccurate images. To solve this problem we have further trained the AI, using a technique called ‘Image to image’ ,meaning bringing the AI images in the expected style for each different type of planet expected, in order to fine tune the wanted result and style of images. In addition we wrote a python code connecting to NASA exo-planets archive looking for the name of the planet entered and his known parameters to understand how the planet is believed to look like ,and returning the fitting text to the specific type of planet.
The project has a number of benefits, first of all it has to potential to give us the ability to illustrate ,based on the latest scientific knowledge ,thousands of exo-planets visuals easily and quickly, compared to today where to get an accurate illustration you have to bring and use scientist time to explain to the artist costing valuable time and money that can be used for research. In edition the project has an education value, letting curious kids an easy and fun access to exploring the exoplanets, and learning what they are and how can we understand how they look and a little about the physics that help us discover it.
In the next part we will go briefly into more details on how we can understand the way an exo-planet probably looks like without seeing any pictures of it. We do that by using physics. For example we are using the mass of a planet ,and compare it to known critical masses so we can classify the planet as a gas rich planet or transitional planet or rocky planet, shown nicely in fig(1).
fig(1) - planet classification by mass-time
where for simplicity we have not entered in-between states, using only the three main classification, meaning rock/icy planets, transitional planets and gas-rich planets(Jupiters and Neptues like). Similarly we can use the mass and temperature to understand the atmosphere expected composition as shown in fig(2)
fig(2) atmosphere composition mass - temperature
which in addition can tell us which planet might have liquid water. Planets that have H2O in their atmosphere that are in the habitable zone, a zone that is defined as the distance a planet needs to be from its star, that makes it possible to have liquid water. It’s to be noted that we assumed an atmospheric pressure of 1atm(earth pressure).
Some other parameters were more straightforward such as temperature, which can clue us in on the planet’s appearance (icy, volcanic, barren,etc…).
Some of the parameters are not as straightforward and need some basic calculations, for example in order to find the star’s color , we used the star’s temperature and wien’s equation, =28981Ts , where Ts is the temperature of the star in Kelvin , and is the wavelength in micrometers(10-6m). Another example is finding the brightness as seen on the planet. to do so we had to find the irradiance on the planet as a result of the star by calculating the irradiance on the planet, f = BTs4Rs2d2[W/m2] , where B is Stefan-Boltzman constant, Rs the radius of the star, d the distance between the star and the planet, and f the irradiance. Then from the irradiance it’s possible to find the apparent magnitude using: mBol=-2.5log(f)-18.997351. The apparent magnitude can tell us how bright the star seems.
The references and the github page of the project can be found in our Credits and resources page.
what is to come?
Moving forward we plan to add much more, starting with refining our AI even more to get better results, in addition we will add the option to create your own planet ,by entering his name and the different parameters needed and seeing him, and later implementing more parameters and different options to generate images of plants ,creatures ,and more planet types to more truly represent planets throughout their life(transition phases as seen in fig(1)). We also want to add the option to choose specific terrain, i.e: a desert area ,a rocky mountain , a valley , an ocean and so on.
some more technical details:
the coded language used for the AI was python, and we used AWS as the training machine, and a gpu machine to run the stable diffusion and python prompt making code.
for the website we have used GoDaddy for the domain and website server, the website was made using Wordpress, html, css ,Javascript.
Important-
because the project ran on a paid server for the time of the hackathon the generate the images after the ending of the hackathon the server will be shut down, meaning it won't be possible to generate the images from the web interface ,and will be possible only via setting up the project locally and creating the pic.
it's important to note the full frontend and backend codes are available in our GitHub in the resource tab, and the web itself is still up and functioning other wise.
we have added a few examples for different exoplanets as a zip file, the name of the planets is shown in the image name, and theoretically it's possible to run and make an illustration for each and every planet with enough available date from nasa exoplanet archive, more specifically from 'Planetary systems composite data' table.
In addition an example of the working site, and image generator in work are added in the end of the pitch presentation, that's available via google drive link in the demo.
some of the examples:
Kepler-1006 b
Kepler-1377 b -
TRAPPIST-1 e
For any question, inquiries you are more then welcome to send an email to:
exovision123@gmail.com
Space Agency Data
We use the NASA exoplanet archive to gather data about exoplanets, specifically from the 'Planetary systems composite data' table. We used this data to extrapolate features of a planet, and then use those features to write a description of the planet and it's star.
Hackathon Journey
Going into the challenge we had a general idea of what we want to do, but not so much how to do so, as for everyone but one of the team had no prior practical experience coding , nor in hackathon or working together as a team. As we started working, splitting assignments, and each person started working, we have encountered a problem after problem, and little by little, step by step started solving them. Starting out without an idea how to make a website, to learning basic html, javascript and css combined with Wordpress, to making a working website, not knowing how we would make the needed dictionary for the learning machine, to reading scientific papers about exoplanets, understanding them and how they are built making each setting of the planet one step at a time until we had a complete dictionary and idea how the planet should look like, all while everyone helped wherever they could to the other teammates if possible. In addition as we kept going we understood we won't have enough time to make everything we wanted in the time set for us, so we sat down at the end of the first day and decided what we are going to do and what we are going to have set on hold for the future. Luckily we at least managed to get a working product, connecting to the the NASA exoplanets system database, getting out all the needed information, understanding the 4 main types of planets, and getting an artput(artistic output) of the landscape we imagine there. We have seen the potential in using AI to create scientific accurate representations of planets, a thing that just a few years ago was unimaginable, and only in 1 day and a half!
We would like to thank the mentors and judges of the local physical event for helping us during the challenge, giving us tips and feedback, while working on the project and the pitch, and thank the monkeytech team for helping when one of our pcs died, finding us a temporary laptop to use, and also for the lectures, food, and place to work during the challenge.
Important-
Because the project ran on a paid server for the time of the hackathon to generate the images, after the ending of the hackathon the server will be shut down, meaning it won't be possible to generate the images from the web interface. It will only be possible via setting up the project locally and creating the pic.
It's important to note the full frontend and backend codes are available in our GitHub in the resource tab, and the web itself is still up and functioning otherwise.
We have added a few examples for different exoplanets as a zip file. The names of the planets are shown in the image name, and theoretically it's possible to run and make an illustration for each and every planet with enough available data from NASA exoplanet archive, more specifically from 'Planetary systems composite data' table.
In addition an example of the working site, and image generator in work are added in the pitch presentation, that's available via google drive link in the demo.
For any question, inquiries you are more then welcome to send an email to:
exovision123@gmail.com
References
- https://en.wikipedia.org/wiki/Circumstellar_habitable_zone
- https://en.wikipedia.org/wiki/Apparent_magnitude
- https://www.astronomy.ohio-state.edu/ryden.1/ast162_2/notes7.html - star brightness
- http://panoptesv.com/SciFi/ColorsOfAlienWorlds/AlienFields.php - plants color
- https://www.researchgate.net/publication/258785366_Relationship_between_Luminosity_Irradiance_and_Temperature_of_star_on_the_orbital_parameters_of_exoplanets
- https://www.nasa.gov/feature/jpl/the-art-of-exoplanets
- https://hal.archives-ouvertes.fr/hal-01877019/document
- https://www.e-education.psu.edu/astro801/content/l12_p4.html
- http://www.astro.sunysb.edu/fwalter/AST101/habzone.html
- https://www.astro.umd.edu/~miller/teaching/astr380f09/lecture14.pdf
- https://www.lpi.usra.edu/education/explore/our_place/hab_ref_table.pdf
- https://geo.libretexts.org/Courses/University_of_California_Davis/GEL_056%3A_Introduction_to_Geophysics/Geophysics_is_everywhere_in_geology.../03%3A_Planetary_Geophysics/3.01%3A_Orbital_Mechanics
- https://www.iau.org/static/resolutions/IAU2015_English.pdf
- https://www.nasa.gov/image-feature/goddard/2022/nasa-s-webb-reveals-cosmic-cliffs-glittering-landscape-of-star-birth
- https://news.uchicago.edu/story/surprise-finding-suggests-water-worlds-are-more-common-we-thought
- https://github.com/CompVis/stable-diffusion
- https://github.com/tsarpaul/nasa-challenge
Tags
#art#web#artofourworlds#ai#machinelearning#exovision#exoplanet#exoplanets#nasadatabase#dataanalys#physics