About this project
Netguru worked with Potara to create an ML algorithm for merging two NFT images into one. This was a novel problem to which a solution had not been developed yet. As a result of the cooperation, the client received a PoC with output samples to attract funding for platform development.
Potara is a platform that allows holders of any two unique NFTs to merge them into one. The two merged NFTs are effectively burned to mint a brand-new token. The owners get different versions of the tokens to choose from, each with distinct visual properties. The newly-created works of art are then subjected to community voting to pick the best one.
Applying Machine Learning expertise to deliver a PoCWhen Potara approached Netguru with their experimental idea for an NFT merging platform, they were seeking machine learning expertise to help bring their vision to life. The challenge they faced was that, at the time, this was a novel computer vision problem, and there was no existing solution to draw upon. The client was looking to obtain a proof of concept for their proposed merging platform, which they could then use to acquire funding from potential investors. Netguru's team of experts worked closely with Potara to develop a comprehensive solution, leveraging machine learning algorithms to create a platform that could accurately merge NFTs while preserving their unique characteristics. Through extensive research and experimentation, the team was able to achieve a successful proof of concept, paving the way for the client to move forward with their plans for the platform.
More specifically, the client expected Netguru to:
- Propose a solution that would provide varied outputs to be used for community voting.
- Conduct in-depth research and experimentation to find the best potential solution.
- Take full ownership of the project, from research, setting up the infrastructure for ML experiments, to development.
- To deliver on the client’s request, Netguru provided comprehensive machine learning services via a dedicated project team composed of: Senior Machine Learning Engineer, Machine Learning Engineer, Project Manager, Account Manager and Quality Assurance Engineer.
Our approach to the project
The Netguru team stayed in close contact with the Potara team, providing weekly updates on the research results. The project began with a review of state-of-the-art computer vision methods for transferring visual properties between images.
Next, the team singled out two possible directions for the solution: semantic style transfer and generative adversarial networks. Style transfer won, as it required shorter iteration time and offered larger variability of NFT images.
The team implemented the pipeline and conducted extensive experiments by merging images from different collections of NFTs and tracking the results in Neptune.ai.
The biggest challenges were:
- Considering various approaches for accounting for the wide variability of NFTs, and ultimately deciding on an optimization framework over model training due to its flexibility and confidence in feasibility. This method allowed for more personalized and tailored results based on individual NFT characteristics.
- Developing a mechanism that could synthesize high-quality textures, ultimately achieving artistically compelling results. This involved implementing a comprehensive process that analyzed and enhanced various features of generated images, resulting in a heightened level of detail and realism. The team's efforts were focused on creating a solution that could generate textures that were not only visually appealing but also functional and useful in a range of applications.
- Potara received a PoC for a merging algorithm that fully supported their vision.
- The client received a detailed overview of the solution, as well as samples to add to a business plan and attract investors.
- The solution was delivered in 6 weeks.
- Netguru took end-to-end care of the PoC delivery process.