CASE STUDY
Running Data and ML services on a stored image
Viam allows Data Management and Machine Learning capabilities allowing you to access data captured by the robot's components.
Viam natively supports TensorFlow Lite ML models where the user is able to train their own image classification models on App or add an existing model for object detection and classification within the platform using data from the data management service.
This project was done in cross collaboration with the Viam Data/ML engineers and is deployed on App.
laying the ground work
Understanding the current and desired information architecture
The current information architecture is confusing.
Viam App separates their Data Management services into two tabs: Data and Models.
On the current Models page, there are two sources of truth: information & actions
current data>models page
🌐 Information
Table of models uploaded including:
name
date created
size
Table of current training models
name
date created
status
duration
any error codes
ACTIONS
Add new model
Select .tflite mode file
Select .txt label file
Add a new version of a previously uploaded model
Select uploaded model
Select .tflite mode file
Select .txt label file
Enhancing the Information Architecture
We can enhance the information architecture by creating the actions as CTAs and display the necessary information directly on the Models page
laying the ground work pt 2
Larger screen for labeling
The app currently only displays bounding boxes at a resolution size of 400x300, which makes the experience of drawing bounding boxes extremely difficult as they need to zoom in on their desktop to get a detailed enough view.
USER PROBLEM
The biologists working at a farm tech company utilizes the Data Management tool to draw bounding boxes around open stomata on cell images.
Updated sidebar view on Data Management screen to include an 'expand screen' option
Expanded screen view for larger screen for bounding box labeling
design ideation
Wireframes passed to developers
Enhancing the Information Architecture
Larger bounding box for labeling
Running a model on a stored image
Last updated on July 20, 2024, 5:46 PM EST