-
Notifications
You must be signed in to change notification settings - Fork 1
Proposal
Project: Night to Light Photo Conversion
Project Time-frame: 10 September 2018 to 10 December 2018
Summary: A web app that lets users upload dark, night time photos and convert them to bright, well lit photos.
Young people enjoy taking photos for social media. Young people also spend a lot of time outside at night to attend bars, clubs, or parties. There is an overlap between photography and low lit, night settings. At the same time, a recent paper has shown that a new machine learning algorithm can convert nearly pure-black night photos into high quality bright photos with little artifacts.
While it is easy to take high quality photos in the day, night photos are often too dark to see subjects clearly. Taking high quality night photos requires an advanced camera that can handle long exposure techniques, a feature that most mobile phone cameras do not possess. People have a desire to take night photos were subjects are clearly visible and the overall photo is aesthetically pleasing. Yet it is impractical to carry around an expensive, relatively cumbersome, advanced DSLR camera.
Of course, cameras come with flash. However, this is not always desired as a flash is highly obtrusive and hurts people’s eyes.
There are existing tools and tricks to convert low light photos into day implemented in advanced software tools such as Photoshop. However, this takes a level of expertise and time-consumption that is not convenient for the everyday user.
Some companies have begun implementing AI-based low light photo edits in their phone camera applications. However, these AI implementations are restricted to the company’s specific phone as a competitive marketing aspect.
These current approaches only fit the needs of a small group of people and not extensible to a larger audience. Why is this problem worth solving or worth solving better?
Since a convenient dark to light application does not exist, there is a gap in the market. Improving access for night-to-light photo conversion would greatly increase customer satisfaction. This technology would help expand this ability to everyday users. The benefit to us would be adding this project to our portfolios.
The target algorithm (from the paper Learning to See in the Dark) purportedly performs better at night-to-light conversions. In addition, our application will bring user convenience and a simpler interface offered by no other existing product. All the user has to do is navigate to the website on their computer or phone and upload a photo to convert. Then after the conversion the user can share or download the photo.
The following pages provide additional background and motivation: “Learning to See in the Dark” Paper GitHub Code Base With Algorithm and Models
What is the goal of this project?
This project will produce a versatile web application, accessible from mobile or desktop, that allows users to convert dark night photos into well lit photos. What are the defining features and benefits of this product? Allow the creation of well-lit night photos without the use of flash. Implementation of a new algorithm in a convenient user application. Ease of interface. The application will responsively adapt to mobile phones and desktops for the best user experience. Sharing with friends on social media websites through company APIs.
We want to focus on a user experience that is easy to use for a large audience of people so the design of the web application will initially be targeted towards amateur computer users for the photo enhancement web-app.
- Designing an easy-to-use web-application that most people can use
- Give the user the ability to pick a photo from their File Explorer
- Give the user the ability to drag and drop pictures from their library onto the Web-App
- The app should be usable from phones
- Building a responsive Web App in a JavaScript/TypeScript framework
- Designing an easy to use interface with good sharing functionality
- Dark to Light Photo-quality enhancement by using newly published Machine Learning algorithms
- Making a corresponding native phone-app
- Designing an advanced photo-editing interface
- Developing new algorithms from scratch
- Early inside-betas for user-testing
- Usable final web application
- Open source codebase on GitHub
- Written tutorial with pictures on how to use the application
- Screenshots of application features
- Documentation of the application
What are the main risks of this project?
There are significant technical difficulties in implementing the machine learning algorithm. This will be a risk because one person on our team has much experience with the relevant tools and technologies. Although the others will learn, we will certainly make some mistakes and suboptimal choices. We will address this risk by scoping the project such that we have enough time to train and to review the design and implementation.
The schedule for this project is very short. We will manage this by planning a conservatively scoped functional core and series of functional enhancements that can be individually slipped to later releases if needed. The machine learning algorithm does not work as well as claimed. If this happens, the risk is not too significant as the main goals of this project is the learning experience. We would still learn how to integrate the algorithm even if the algorithm itself does not perform well.
There can be risks in designing a Web-App that provides a good user experience, because it can be hard for Software Developers to design applications for most people. We will be communicating, maybe surveying with our target audience to address whether they or not they have concerns or improvements to our Design and User Experience.
We want to keep our application ad-free and open source; thus the main rewards will be the learning experience from doing a group project in an Agile framework, alongside learning more about Machine Learning implementation, and earning a good grade on the project and the class.