213COMPENSO

https://github.com/dogruaar/spaceappschallenge/blob/main/213COMPENSO_Space_Apps_Challenge.ipynb

https://drive.google.com/drive/u/2/folders/1aLzxL6F8h_JzxASIPL-JtOTC0fitarYo

The main purpose of the project is to process the photographs taken by NASA with artificial intelligence and present them to the user in artistic way. The project creates AI arts from the NASA's Images Library. It uses Difussion algorithm in Machine Learning. Basically, Diffusion Models work by destroying the training data by cascading Gaussian noise and then learning to recover the data by reversing this noise process. After training, Diffusion Model can be use to generate data by passing randomly sampled noise through the learned noise removal process.

Nowadays, people see a lot of things on social media, on the street, and pass by without focusing on any of them. In order to prevent this, we elaborate ordinary photographs and we develop them in such a way that people want to stop and look and think for a while.

Instead of building an AI search engine, we used NASA's search API. The reason of this, it was impossible to create such a large dataset that we could train AI in 24 hours. Also, training an AI with this dataset would take days.

How does it work?

The project creates an artistic output by analyzing the images taken from the NASA database with AI and Machine Learning methods and processing them according to the user's request. To explain step by step:

1. Take an input from the user for space images
2. Search the input in NASA Images
3. Analyze images
4. Ask to user for choosing one of these images
5. Get an artistic prompt from the user
6. Run the algorithm
7. OUTPUT!

An example:

Input:

Output:

ACHIEVING THE GOAL

This project was created in accordance with its purpose, to attract the attention of the public. Space, which has inspired great artists and scientists until today, will now inspire all humanity, whether they have a talent for painting or not. The freedom that the application gives to the user will make each person a space explorer. They can discover, learn and produce at the same time.

For achieving this goal we used Python on Google Colab Notebook. Also, we have used a lot of third party libraries are like Pytorch, numpy, pandas etc . We have rented an A6000, which is a graphich card with high VRAM, from vast.ai.

We have used some open source projects mentioned in list:

• Open AI's Clip
• Clip Interrogator
• Katherine Crowson's fine-tuned 512x512 diffusion model
• Disco Difussion

For creating arts from photograps we choosed the NASA's Images Librarybecause of its rich content.

Thanks to this hackathon, we learned to work efficiently and to be creative in a short time. We learned to work as a team with our teammates, to solve each other's problems and to teach each other new ideas at the same time. We were very interested in the idea of creating a work of art with words, and thanks to this idea, we developed this project. We had to complete this project in a short time, so we were expected to develop fast and practical solutions to the problems we encountered. Despite all the problems we experienced, we completed our project with fast and practical solutions without giving up. In addition, we were assisted by the experts in our work environment while solving the problems we encountered. In this way, we learned how to solve our problems from different perspectives. We thank I3P for their hospitality and working environment.

1. NASA Image and Video Library
2. Open AI's CLIP
3. CLIP Interrogator
4. Katherine Crowson's fine-tuned 512x512 diffusion model
5. Disco Difussion

#AIArt, #Art, #Diffusion, #ML, #AI