Forxai platform logo

Appearance

What's New in 3.0 ​

Welcome to FORXAI Video Vision 3.0!

Summary ​

This major release brings support of NVIDIA 5xxx series and adds dataset and annotation handling, which are foundations for future model training functionality.

Following major functions were added in this release:

With this release several breaking changes are introduced, review them for your reference when upgrading from older versions:

See information about hotfix updates:

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.

Enhanced AI model support and Model Management ​

We've significantly improved our AI capabilities and model management:

  • Local model management in HUB: Enhanced model management capabilities directly within the HUB interface, providing better control and organization of AI models. (Models).
  • NVIDIA 50xx GPU Support: Added support for anomaly detection and object detection model inference on the latest NVIDIA 50xx GPUs using Triton Inference Server, delivering improved performance and efficiency

Dataset annotations ​

New data management features to improve workflow efficiency:

  • CVAT Annotations Support: The platform now supports storing CVAT annotations, enabling seamless integration with computer vision annotation tools for better data labeling workflows in datasets. (Annotations)
  • Active Disk Checking: Implemented active disk space monitoring when uploading files into the platform, ensuring sufficient storage availability and preventing upload failures

Builder Enhancements ​

Improved development and testing capabilities:

  • Camera Device Preview and status (Camera Healthcheck): Added real-time camera device output preview functionality in the builder, allowing developers to visualize camera feeds and test configurations more effectively. (Flow Builder)

Compatibility ​

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

This version powers minor release of Forxai Mirror v2.7.0.

Breaking changes ​

Changes below have an impact on already established functionality and may cause disruptions when updating existing production deployments to newer versions.

Crop Image Node and Polygon Threshold Node Backward Compatibility ​

There is a known backward compatibility issue with the Crop Image node and Polygon Threshold node in this release.

Prior versions of these nodes had configuration fields with percentage ranges from 0.0 - 1.0, meanwhile the new versions of these nodes use ranges 0 - 100. Due to these changes in their configurations, existing flows using this node may not function correctly after upgrading.

To resolve this, please remove any existing Crop Image nodes and Polygon Threshold nodes in your flows, then add and configure them again. (Nodes)

Torchserving inference server ​

With this major release official support for torchserving inference server is dropped.

Existing functionality is replaced by triton inference server, with support for additional problem types and model training planned for future releases. This makes past projects with models trained for torchserving backend incompatible with this version, meaning in case of updates to this version re-training of those models is needed.

Grafana dashboards by default ​

Grafana has been removed from the standard deployment of FVV. It can still be activated, if necessary, by specifying 'grafana' as an additional docker compose profile, in addition to the fvv-platform-gpu profile.

Changes in hotfix 3.0.1 ​

Issue with passing images from the virtual image feeder to the deployment in the wrong format, which caused significant NN model performance degradation.

Changes in hotfix 3.0.2 ​

Issue with deployment freezing due to a camera device not reconnecting after connection issues.