Ultralytics
Ultralytics is a company that allows users to create and train AI models. Users can create them for their businesses or projects or integrate them in their apps. The solutions include healthcare, self-driving cars, agriculture, sports and more.
PRODUCT
INDUSTRY
Mobile app
MY ROLE
Product designer
AI
DATE
2023
Goal
Our goal was to redesign Ultralytics app, add more functionality and create UI that can be later used across all Ultralytics platforms – web app, website and mobile app.
Target audience
How it works
In the mobile app, users first upload a dataset or use the pre-uploaded ones. Then they train their AI models on them. They can also create projects that consist of one or more models.
Initial structure
Initially, the mobile app had very limited functionality. Users could only test their models there. Our goal was to adapt web features to the mobile version, add new finctionality and significantly improve the UI.
New structure
In the new version of the mobile app, users can upload datasets, train models and create projects as well as share and save them.
User flow
Onboarding
Predict
Predict tab is camera where users can switch modes (Pose, Detect, Classify), select models they would like to test and apply filters to get the result they need.
Pose, Detect, Classify modes
When users open the app for the first time, they go through a quick onboarding session that allows them to understand what they can do with the app
Upload a photo
Users can either use their camera or upload a picture from their gallery to test it.
Search for a model
Users can select a model they would like to test. Models can be public or private. Some public models are available only on Hub Pro plan. Private models are created by users.
Datasets
Datasets contain images that users can use to train their AI models. Datasets are also divided into public and private. Users can't upload datasets on Hub Mobile, they can only upload them on Hub Web.
Dataset page
The dataset page contains all the information about a dataset, including classes and projects.
By viewing classes users can understand what this dataset is about without scrolling through multiple of images.
Users can see what projects and models this dataset belongs to. They can go directly to projects and models from here.
Statistics shows distribution of images into 3 stages: training, test, validation.
The same information is shown if the user presses on a class name.
Projects
Projects contain several models. They can also be private and public. The users can't create projects here. They are created on the Hub Web
Project page
Users select models they can see on the metrics graph. It helps to filter down most used models, as there can be dozens of models in one project.
On a project page the user can see metrics and models this project consists of.
Metrics graph
One of the tasks from the client was to redesign the metrics graph for mobile, as it was too large and difficult to read.
Metrics graph v.1
We decided to get rid of the graph with lines and use bars instead to allow the users to see the state of the models at any given epoch.
If the user has chosen fewer models, the bars become wider.
When they press on the epoch, they can see the state of models at this stage.
Metrics graph v.2
We tested this graph on a bunch of users and it turned out it had a couple of disadvantages.
Metrics graph v.2
We changed the graph to make it more like the graph in the desktop version and easier to scroll.
The version that was approved
The version we discarded
Model page
From the project page users can go to the model page. Users can download models from here in different formats, share it or see the training results
Profile
The profile page is quite simple. It contains plan information and support.
Please open this page on desktop to get better experience.
Ultralytics website
We also designed a website for Ultralytics with a matching UI and branding.
