PixelVerse

The backstory and our motivation

We were really excited about this hackathon and we had a lot of cool ideas, most of those ideas were around training our own vision model. But our excitement was short lived because we were able to source necessary data sets to train our own model. We could have used any data to train our model over the weekend, but here's the kicker: the effectiveness of these models hinges on the quality of the data they're trained on. It's like cooking—a gourmet meal depends on top-notch ingredients. Quality always trumps quantity.

Sourcing data is the cornerstone of effective machine learning models. According to a report by Forbes, 80% of AI projects never make it out of the lab due to data issues (Mar, 2019). Another study by MIT Technology Review Insights found that 72% of companies experienced data quality issues that impacted their AI and machine learning projects (MIT Technology Review Insights, 2021).

Currently, acquiring high-quality data poses significant challenges. One way to obtain data is by creating your own dataset, which involves manually creating images—a time-consuming and labor-intensive process. Alternatively, you can source data from popular platforms providers. However, even these platforms often fall short when it comes to highly specific data needs.

TLDR; of what we offer

PixelVerse addresses this gap that we just talked about in the sections above by providing a platform where users can request highly specific datasets, which we call "projects." 

Users can now receive data contributions from the public tailored to their exact requirements. For a user, it's as simple as requesting a dataset, putting in some sample images for the contributors to follow. The public, in turn, can contribute images to these projects and earn token points.

Users are incentives🤑 to contribute because they earn our tokens which they can either use to redeem gift cards or make new projects for them to source images.

Essentially, our product addresses the issue of sourcing quality images, which results in subpar models. We understand the critical importance of high-quality data and have developed a sustainable solution that offers the public access to this powerful tool and open source support.

BAD Actors: Whenever a photo is uploaded we run it through a couple of checks in our pipeline. This is were gemini comes in, we use it in conjunction with the sample images uploaded so we can sure that the photo is relevant

Technologies Used

Challenges faced and what we learnt from them

• FRONTEND: We are all backend developers who are proficient in many technologies but front end. We breezed through the database and microservice aspects of our project but hit a wall with the frontend. It was so bad that we had to reset our front end repo on the second day because of bad picks on UI libraries.
• INTEGRATION HELL: It was very complicated to integrate our back end and front end services, especially when we are using different providers for different tasks.
• Animation (do I need to say more)

We learned how important is good research on UI libraries before jumping the gun because you will eventually you will run into problems. We also learned a lot about front end development and best pratices to avoid problems.