Final Pitch Deck
Individual Self-Reflections
Prototypes
Role
People have busy lives – we know. It’s hard to take time to slow down and step back and reflect and relax and remember everything that’s going on amidst the chaos. We also realize that taking the time to do so is something that many people simply cannot afford to do with their hectic schedules. We’re here to make that easy and simple and painless. With this lived reality in mind, we thought of how to create an experience with Memento that would positively impact our users’ lives.
We envision our product as a lightweight daily perk, where they interface with the app through whimsical and thoughtful daily prompts that lead them to reflect intentionally on what they’re doing and stop and appreciate not only the highlights and “big” events, but also all the little moments in their daily lives as well, such as a fulfilling conversation with an friend or a shared meal or even a single thing that made them smile. We hope that these daily pings won’t be regarded as another “to-do” list item to check off on their daily routines, and instead be something they look forward to receiving every day, like a daily challenge or scavenger-hunt-esque task. The culmination of these daily action items (which simply involve them submitting a picture and response), is a physical scrapbook photo album of their past month, which we hope will evoke fond feelings of nostalgia and reflectiveness and warmth and joy at seeing the collection of all the moments, big and small.
This product is meant to serve not only as a physical keepsake, as something that can be affectionately looked back upon for reminiscence or sharing stories and memories and bonding with friends, but also as a means for reflection of our users’ daily lives where they can see the progression of everything they’ve done and accomplished and gone through, and also inspire and motivate them to live their daily lives with thoughtfulness and intentionality and to realize and appreciate that their lives are so beautiful and deserve to be celebrated and that always, Every Moment Matters.
Look & Feel
The goal of our product is to evoke a sense of warmth and welcoming as well as inspire excitement and energy. We realize that daily prompts may seem like a real commitment and require user action, so we want to make the experience as lighthearted, lightweight, and interesting as possible. To this end, we utilize creative and unique prompts daily, coupled with bright primary colors for a more playful feel when interfacing with the app. These bold, primary colors also evoke nostalgia as a nod to childhood. The contrast between the popping colors in the palette and their brightness is an attempt to stimulate and rejuvenate users and spark joy in the whimsicality of these hues – the choice of fonts is for simplicity and a clean, modern touch. We want our app to be clear and straight to the point, but also refreshing to use.
Implementation
Videos
A major technical challenge is the process of implementing ML photo recognition and organization. The aim of the ML algorithm is to categorize photos into groups that would look well together on a scrapbook page. The algorithm will first detect faces and group photos on whether they contain people in them or not. Then, it will go through all of the photos with people in them and group together photos that have similar people in them. It will also go through all of the photos without people and group together photos that have visually similar content.
Our implementation will involve the use of the OpenCV and dlib libraries. OpenCV is an image processing library, and dilb is a machine learning library. Both are extensively used for facial recognition tasks. We will use the photos that the users have taken as input. These photos will be preprocessed using techniques such as scaling, transforming, and data augmentation. The purpose of these techniques is to add more examples to the dataset to make the dataset more robust. We will also leverage the results from pre-trained models such as the public face recognition datasets FaceScrub and CASIA-WebFace in addition to the data we collect from users.
First, we will perform face recognition to detect the faces in our photos. To do this, we will use the dlib library and the face_recognition repository. We will first detect and segment faces. We will do this by using face_recognition.face_landmarks(). Then, we will compute embeddings to quantify the faces and train a Support Vector Machine (SVM) with the face embeddings. Finally, we will use our model to recognize the faces.
After recognizing the faces, we will build another model to cluster photos with similar groups of faces. To do this, we will represent each photo in vector format with embeddings representing the identity of faces in the photos. Then, we will use the k-means clustering algorithm to partition the photos into k groups, where k corresponds to the number of scrapbook pages. In order to implement this, we will use the sklearn.cluster.KMeans model from the sklearn library.
We will then work with the photos without people in them and group together photos with visually similar content. After loading the data, we will use the VGG16_BN model from Pytorch Torchvision. This is a deep CNN that uses batch normalization in order to reduce covariate shift. We will use the pre-trained weights and parameters for this model. Then, we will use k-means clustering to group the images. Similar to the process for grouping faces, we will use the sklearn.cluster.KMeans model from the sklearn library. The number of clusters will be based on the number of scrapbook pages for the user.
Finally, we will build scrapbook pages based on the photos that have been clustered together. The process described above will help in the organization of scrapbook pages by building pages that feature similar groups of people in the photos.
