Forxai platform logo

Appearance

What's New in 3.2 ​

Welcome to FORXAI Video Vision 3.2.0!

Summary ​

This minor release brings significant enhancements to the platform, focusing mostly on implementing model training to the platform. Let's explore what's new!

With this release documentation was overhauled to bring more structure to new users.

  • Getting started section is aimed at new users to provide them with learning path, gradually introducing them to the product.
  • How to guides are aimed at users familiarized with basic concept and provide overview steps to accomplish specific task.
  • Visit References section for detailed description of specific functionality and detailed description of behavior.
  • Release notes are now a separate page, with overview of past releases available.
  • Tutorials are useful for users who want to learn new skill and provide clear journey for learning.

Other major functions that were added in this release:

See information about hotfix updates:

Launch of Self-served Model Training ​

With this release we are introducing capabilities for the user to train their own models and deploy them for use within configured Projects.

Your journey starts by defining their workspace, which is a container for model training problems to organize them. In model problem, you can configure your experiments. During experiment configuration you can tune several settings for model training - datasets, preprocessing information, augmentation and hyper-parameters. When finished you are ready to start Model Training and see its progress. You are notified about status updates and presented with metrics for model evaluation. Models trained during model training can be downloaded, uploaded to central model storage and used in Projects once deployed.

With launch of this feature we are adding support for training object detection and anomaly detection models, with more problem types and supported neural networks planned for future.

Major Features ​

Major features introduced in this release are described below. This release includes other changes related to fine-tuning FVV capabilities, bugfixes, UX and other improvements which are omitted in this article.

Model management enhancement ​

This release adds support for downloading models locally from the Hub and Builder, along with a simplified way to add models from the local store directly into a project. Read more about this function in dedicated section for Local Models

Extension for polygon labels ​

Annotate Polygons node has been improved with an optional custom labels which outputs label string together with model probability. Also new integrated color picker tool has been added, allowing for intuitive customization of polygon and label color directly within the node interface.

Access control for central model storage ​

Starting this release users will be able to download models based on access rights of account used during installation of FORXAI Video Vision. If you are unable to access models you need, contact your local Konica Minolta representative to support you with resolving the issue.

Notable fixes ​

Fixes for major issues impacting deployments in production are noted below.

Dashboard connectivity issues ​

Connection of dashboards to Flows was reinforced and is now more durable. Additionally, this connection is actively refreshed behind the scenes to avoid its expiration resulting in requiring users to sign in.

Third party licenses on dashboard page ​

To eliminate risk of getting stuck on license page or navigating there by error, link to page with third party licenses was removed from Dashboard UI. If you are interested in viewing licenses of software used, they are available in Hub application.

Report generation fails with bigger data sizes ​

We eliminated some overhead and optimized report generation to ensure that reporting generation does not fail for bigger reports as previously discovered. Users generating bigger exports (more than 5M records) are still advised that downloads may take time due to the sheer amount of data included in such reports.

Improved browsing speed in reporting ​

We have updated the reporting viewer to make browsing through your records faster and more reliable. You can now scroll through the most recent 10,000 records directly in your browser. This covers the vast majority of daily monitoring needs while keeping the interface smooth and responsive. If you need to analyze your full history beyond the 10,000-record preview, simply use the Export button to download a complete report.

User notifications ​

Users are able to review important actions that take place by observing notifications. These inform users when unexpected errors take place, or when progress or state of longer actions changes such as model training or generation of exports. Read more in documentation for notifications.

Compatibility ​

The Video Vision platform was tested on Ubuntu Server 24.04 LTS.

Known issues ​

  • External connection triggers missed: If trigger to external connections is sent, while FORXAI Video Vision is starting, it may not be processed properly. To ensure that external connections behave as expected, allow complete startup sequence to take place.
  • Adding inputs/outputs to Flows requires remapping deployment: When new inputs and outputs are used in Flows, these new inputs are not shown when setting Deployment mapping in Deployment Editor. To be able to see them, users are required to delete mapping for dedicated Flow and recreating it again.

Breaking changes ​

  • Annotate Polygons: Due to the introduction of the Color Picker, new node input and updated node configuration options, existing Annotate polygons nodes are not backward compatible with their previously saved settings. To ensure correct behavior, replace any old nodes in your flows with the newly improved version and reconfigure their parameters.
  • Heatmap Threshold: The output port has been renamed from thresholded_heatmap to thresholded. A new Binarize heatmap parameter (default: enabled) now converts above-threshold pixels to 255 instead of preserving their original intensity. Flows connected to the old output port name will need to be reconnected. To ensure correct behavior, replace any old nodes in your flows with the newly improved version and reconfigure their parameters.
  • Overlay Heatmap: A new Display zero values parameter (default: disabled) controls whether zero-intensity heatmap regions receive the color overlay. The default behavior has changed. Only non-zero areas are colored, but flows that previously relied on full-image overlay should enable this parameter explicitly. To ensure correct behavior, replace any old nodes in your flows with the newly improved version and reconfigure their parameters.

Changes in hotfix 3.2.1 ​

Fixed reconnecting issue with Basler cameras in hardware trigger mode.