Encord adds native MCAP and time series labeling, plus new medical imaging tools

MCAP recordings and time series data

According to Encord, robot sensors log at different rates, and frame-based labeling forces them onto one grid. That requires discretization and can make labels drift from the data. Encord now lets teams add MCAP files exactly as they were captured, either by direct upload or from cloud storage.

The company says this removes the need to build an extraction pipeline, or to rewrite one whenever sensors change. Key sensor streams appear on one continuous timeline, even when they start at different times or run at different rates. Labels are tied to the recording's own timestamps, so, according to Encord, there are no offsets to fix before labeled data goes into training.

MCAP support is in beta and switched on per workspace. Customers need to ask their account team to join.

Encord also supports uploading time series data. Some important moments only show up in signals, the company says, such as a spike in joint torque when a grasp fails or a stretch of hard braking on a velocity trace. Users can label these moments as time ranges, alongside the camera and LiDAR data showing what was happening at the time. Account teams can help with setup.

Fewer classes per workflow stage

For projects with long ontologies or workflows that need specialist passes, Encord now lets teams choose which classes each annotation and review stage can edit and which it cannot see at all.

• Focused stages: annotators only see what their stage is responsible for, so there is no scrolling through a long ontology or picking the wrong class.
• Specialists where needed: expert annotators and reviewers can work only on the classes relevant to them, without combing through whole tasks to check a handful of labels.
• Locked earlier labels: later stages cannot change labels for classes outside their scope.

The feature is in beta and rolling out gradually to all teams. Teams that need it sooner can contact their account team.

Upgrades for DICOM and NIfTI labeling

Encord has upgraded DICOM and NIfTI labeling in Projects. Radiologists can view scans in oblique planes with temporal navigation and label 4D NIfTI scans across all axes. They can read tissue density with intensity projections (MIP, MinIP and AIP) and check HU and density values for any selected region. Annotation across MPR views is described as smoother, with mouse-friendly controls and faster image rendering. This part is in beta and enabled on request.

Finding and moving between scans has also changed:

• Teams on Curation or the full platform can filter DICOM and NIfTI data in Explore by metadata, bring their own embeddings to search and filter, and preview real image frames as thumbnails.
• A "Preload Next Task" option in Project settings downloads the next task in the queue in the background, so radiologists do not wait for files to load between tasks.
• Cloud-synced folders now support NIfTI files, so new data can be imported automatically into annotation projects.

Other changes in the release

Encord also listed several smaller improvements:

• Polygons: a polygon with disconnected parts can be split into separate objects (right-click > Split polygon, or Shift+P). Vertices can snap onto neighboring polygons so shared edges line up exactly, with snapping enabled in Settings > Drawing.
• Cuboids: interpolating cuboids in 3D scenes no longer depends on hidden hotkeys or right-click menus. Users turn on Auto-interpolate to use it, which Encord says cuts time on a frequent LiDAR labeling task.
• Issue review: the issue pane has a cleaner layout, threads open only when an issue or its object is selected, and new filters show only issues for the selected object, or only rejections.
• Custom Agents: starting with encord-agents SDK v0.2.10, Encord sends Custom Agents a signed notification when tasks are ready, so they no longer need to poll for work.

What it means for teams

Encord frames the release as making it faster and easier to turn raw data into training-ready data. For robotics teams, the main change is that recordings and signals can be labeled in the format they were captured in, with labels tied to original timestamps. For teams with large ontologies or specialist reviewers, class scoping per stage is meant to reduce errors and wasted effort. Medical imaging teams get tools aimed at faster reading and less waiting between tasks.

Several of the larger features, including MCAP, class scoping and the DICOM and NIfTI upgrades, are in beta and require contacting an account team or waiting for rollout, so availability will vary by customer.