Product Roundup 26.Q1.1

Confusion matrix CSV for QA, more reliable uploads, faster YOLO Classification imports, stable 3D annotation downloads, and key security fixes including an OPA upgrade.

New

Download quality report as a confusion matrix (CSV)‍

You can now export quality reports as a confusion matrix CSV. This makes it easy to analyze misclassifications in spreadsheets, share results with stakeholders, and feed downstream QA dashboards.

Fixes

More reliable uploads and task creation

• TUS uploads now read HTTP request data in chunks instead of loading the full body into memory. This reduces memory pressure during large uploads.
• After a TUS upload completes, HEAD requests now return 404 instead of 500. This prevents false alarms in CI and health checks.
• When creating a task, manifest files uploaded with videos are no longer ignored, so media and metadata stay in sync.

Faster imports and stable exports

• Ultralytics YOLO Classification imports are faster, improving iteration speed for classification workflows.
• Fixed an AssertionError during 3D annotation download, restoring reliable export of 3D work.

Smoother annotation UI

• Removed redundant PATCH /api/jobs/<id>/meta calls from the annotation view to cut unnecessary traffic and reduce noise.

CLI agent resilience

• Fixed agents waiting an inappropriate amount of time between retries when acquiring AR tasks.
• Agents now handle failures of queue-related HTTP requests more robustly, improving long-running automation.

Security

• Fixed XSS vulnerabilities related to skeleton SVG images.
• Upgraded OPA to 1.12.2, aligning with current policy-engine improvements.
• Staff users can no longer change their own permissions, closing a privilege escalation gap.

Have suggestions or requests for what you'd like to see next? Open an issue on GitHub or contact us via Help Desk.