AI App Idea: Crowdsourced Image Dataset

The world’s largest image dataset is probably Google’s JFT-300M which has over 300 million images to date. But it’s meant for Google’s internal use only. The next biggest is ImageNet which contains more than 9 million images with over 16 million bounding boxes. These boxes are boundary lines that confine objects in images used to help AI to identify the objects.

Now, what if there’s a way for anyone to annotate their own images and submit them to a dataset that they can eventually benefit from?

One-Liner Pitch

An image dataset for anyone to submit and annotate their images to train an AI and in return get to use the AI system for free.

The Problem

We need an AI brain to learn all our photos on our mobile phones and help us search when we need to retrieve them. Yes, Google Photos is supposed to solve this problem but Google’s photos dataset isn’t shared with the AI community. It’s private for Google’s own use only. And, Google Photos isn’t that accurate in detecting our photos. Try searching for your photos and soon you’ll realize it missed a lot of your photos.

The Solution

To create a mobile app where users like us could upload curated photos and help annotate the photos for AI learning. In return, we get to use the image annotations created by all of us to help us search through all our photos on our phones. It’s like the Google Photos app, but a lot more accurate and functional.

We are tapping on the wisdom of many to help many. So, This is a win-win app where people help to tell what is contained inside their photos and submit them to the system to help train an AI model which eventually helps us to search through our own photos.

The USP (Unique Selling Proposition)

Concerning copyright, today OpenAI is scraping petabytes of public data from the Internet to train its GPT models, and they aren’t really paying others for it. For now, that seems to be how the AI rules of engagement work. Now, here’s the kicker. However the tide turns in AI law in the near future, this app shall be immune because the app users are submitting their photos in their own free will. This means users are providing a license for the app to use their photos where the copyright of the photos still belongs to the users.

The Caveats

We should not submit our private photos. So, we need to discriminate between what photos we want to annotate and submit and what not. The app eventually should be smart enough to help us distinguish what photos we can submit.

Our Hope

One fine day soon, we hope someone will build this AI solution for the benefit of all.

2 comments

Leave a Reply

Your email address will not be published. Required fields are marked *