Customized Avatar, HUAWEI
HCI Design intern | Beijing
Status
Already launched
My Role
HCI Designer
Tool
Unity, Blender, Figma
Duration
Mar, 2021 - Nov, 2021
My contribution
Conducted two rounds of user surveys, obtained insights into user avatar preferences and developed a deliverable avatar modeling process.
Project overview
Aimed to introduce an Avatar system based on facial feature recognition that aligns with the aesthetic preferences of Chinese users, in order to meet their personalized needs. This feature showcased at HarmonyOS 3.0 release conference, reaching over 200 million HarmonyOS users.
Project Process
In this project, we answer one question: how can we translate a user's vague sense of aesthetics into a deliverable avatar modeling algorithm?
Step1
Obtain users' preferences for Avatar styles
Step2
Transform the user's perception between avatars and real people into deliverable data
Step3
Integrate the data into the algorithm and process for Avatar creation
Step4
UI Design
Step #1
Obtain users' preferences for Avatar styles
We have collected various mainstream Avatar styles on the market and categorized them according to Mature, Childish, Unrealistic, and Realistic. We designed and distributed over 500 surveys and obtained users' preferences for avatar style.
Insight #1
The body-to-head ratio is more realistic
Compared to the cartoon style of FaceBook or Zepeto, where the body and head are smaller, users prefer a style where the body and head are more proportionate.
Insight #2
The facial features is more cartoonish
Compared to Samsung's style, users prefer facial features to be more cartoonish. However, compared to Apple's style, they hope to retain more of their own unique characteristics.
Step #2
Transform user's perception into data
Based on the style we got from the first step, we began to find the answer of how could we map the Avatar to facial recognition data. . We found over 50 participants and collected and summarized their data.
Based on the style we got from the first step, we began to find the answer of how could we map the Avatar to facial recognition data. . We found over 50 participants and collected and summarized their data.
Data
Obtained from user research
From this, we obtained data on the mapping coefficient the Blender Avatar Shape Keys and the facial detection data.
Step #3
Integrate the data into the algorithm and process for avatar creation
Avatar Modeling
Through mapping the mapping coefficient with Blender's Shape Key, we can generate Avatars that match users' facial features and meet their expectations.
Machine Learning Algorithm
The mapping data we obtained can also be applied in the loss function of machine learning. It can be adjusted based on feature weights to ensure that features with higher scores are more accurately represented in the Avatar.
Step #4
UI Design
This feature showcased at HarmonyOS 3.0 release conference, reaching over 200 million HarmonyOS users.
Feature Entrance
This feature is integrated into the Huawei account. Click on the user panel to enter the Huawei Account Center, and then click on 3D Avatar to access the Avatar creation interface.
Feature Page
Click 'Take Selfie', take a selfie, and an Avatar matching the selfie will be generated
Modify
After generating the Avatar, users can further adjust the facial features according to their own preferences. Then, they can choose their favorite clothing to match.
Expand to other scenarios
The avatar can be integrated with various applications, such as off-screen displays, stockers.
