Smartphones now have professional grade cameras and there is an increase in great photos captured by amateurs. Phonix powers such users with the ability to monetize such photos with ease.
What it does
It identifies the most valuable photos and provides an easy way for users to publish these photos to an online marketplace. This unlocks the kudos worthy photos that would have otherwise just end up in their social feeds and might have been soon forgotten. The iOS app runs on the user’s devices in background and once Phonix’s advanced Image recognition engine identifies the right candidates, the user is notified. Each photo has a potential dollar value and a list of relevant tags. Once the user approves the selected photos, it is uploaded to the marketplace. The user gets their share of revenue whenever their photo is bought.
How we built it
Seller platform (iOS application) - Generated scores for images using a simple algorithm in conjunction with Firebase ML SDK to identify image features. These images were then assigned a potential dollar value based on a score to dollar relationship model. The photos can be reviewed, selected and published to firebase realtime datastore backend. Backend - Using firebase realtime datastore as the backend to store the published image information and the number of items sold. Buyer platform (web application) - Built with JAM stack and used Firebase SDK for the real-time data exchange.
Challenges we ran into
*We wanted to integrate with the APIs of existing Stock Photography services. But most of them only support manual uploads and Adobe Stock has such an interface but the adobe account requires a review and approval, so we ended up creating our own web marketplace. *We did not find big enough labeled photography data to train the scoring and pricing models, so the algorithm at this point is simple and intuition based.
Accomplishments that we're proud of
*Built a privacy-conscious app by performing all the image recognition tasks on the device *The Seller-Buyer experience is real-time and showcases the power of this idea effectively
What we learned
*Image recognition technique and tools *Standard image related metadata like EXIF
What's next for Phonix
*Quick high-value feature identification in images like a blur, focus, histogram analysis. Also, training the scoring algorithm based on these new features. *Image filtering to remove repeating images. *Photography project creation based on assigned theme and tags, this will improve the quality of suggestions and will also keep the users engaged.