DeepLabCut-Live-GUI v2.0.0

This release expands hardware synchronization, high-throughput multi-camera recording, frame timestamp reporting, monochrome capture, and custom processor integration.
It also improves data safety, diagnostics, and general usability.

Camera triggering and synchronization

Added

• Hardware trigger configuration for compatible GenTL and Basler cameras
• Trigger roles for common synchronization setups:
  • Off / Free-run for continuous acquisition
  • External trigger for cameras driven by an external hardware signal
  • Follower for cameras synchronized to a master camera or shared trigger source
  • Master output for free-running cameras that emit a synchronization signal
• Configuration for trigger selector, input source, activation condition, and frame read timeout
• Strict mode to reject unsupported trigger configurations instead of applying them best-effort

Changed

• Added a dedicated trigger settings dialog to the camera configuration workflow
• Active camera labels now show configured trigger roles
• Multi-camera startup now prioritizes external and follower cameras before free-running cameras and trigger-generating masters

Fixed

• Improved handling of unsupported or incomplete trigger configurations
• Improved expected timeout handling while cameras wait for hardware trigger pulses
• Improved trigger cleanup when cameras stop or close

Frame timestamps and synchronization data

Added

• Optional hardware frame timestamp reporting for supported cameras
• Basler camera timestamp-counter reporting with raw tick values and tick-frequency metadata when available
• An updated timestamp JSON schema containing:
  • Per-frame software timestamps
  • Optional per-frame hardware timestamp values supplied by cameras or SDKs
  • Timestamp source, timebase, unit, and conversion metadata
  • Explicit indexes for written frames

Changed

• Software and hardware timestamp sources are now described separately
• Timestamp entries are added only after frames are successfully written by the encoder
• Documentation now distinguishes Unix wall-clock timestamps from device-specific camera clock values

Fixed

• Improved timestamp consistency when frames are dropped before encoding
• Preserved raw hardware timestamp values alongside converted values
• Improved fallback behavior when hardware timestamps or conversion metadata are unavailable

Multi-camera performance and recording

Added

• An optional Use faster encoding parameters setting for supported FFmpeg software codecs
  • This can improve encoding throughput for high-resolution or high-frame-rate recordings, at the cost of larger files
• A dedicated full-rate recording path for each active camera
  • Recording is independent of the inference and display pipelines
  • Display or inference slowdowns no longer directly limit recording throughput
• Recorder diagnostics for throughput, write latency, queue depth, and dropped frames

Changed

• Decoupled GUI display from camera capture, pose inference, and recording rates
  • The GUI display is capped at 30 FPS while cameras can continue running at their full rate
  • Inference continues to discard stale frames
• Improved use of each camera's reported runtime frame rate when configuring its recorder
• Moved recording dispatch and frame transformations closer to camera workers to reduce work on the GUI thread
• Reorganized recording logic into dedicated services

Fixed

• Prevented asynchronously queued frames from being modified through reused camera buffers or NumPy views
• Improved frame-size mismatch and encoder error handling
• Improved recording shutdown, retry, cleanup, and UI recovery behavior
• Prevented recording controls from remaining disabled after a failed stop operation

Monochrome camera performance improvements

Added

• A Preserve mono frames option for supported backends
  • This avoids unnecessary conversion of monochrome frames to three-channel color
• Native monochrome frame support for compatible GenTL and Basler cameras
• Camera output-format settings and runtime format reporting
• Camera probing for detected pixel format, emitted frame format, and recommended mono handling

Changed

• Camera labels and diagnostics now indicate preserved mono output
• Improved reporting when the camera pixel format differs from the frame format emitted to the application
• Grayscale conversion remains available in the recorder when required by the selected encoding path

Fixed

• Improved handling of monochrome and higher-bit-depth frames before display and encoding

Custom processor workflows

Added

• Processor hooks for recording start and stop events
  • These allow custom processing, saving, and cleanup to remain synchronized with recording
• Recording context containing session, run folder, filename, and output file information
• Support for processor plugins that must be constructed in the inference worker
• Clearer processor status and recording state information in the GUI

Changed

• Improved custom processor discovery, selection, enablement, and settings persistence
• Processor-controlled recording now uses the standard recording lifecycle
• Stopping inference while recording now provides a clearer warning that processor data may remain unsaved or unpaired, depending on the processor implementation

Fixed

• Improved processor cleanup and finalization when recording stops
• Prevented duplicate recording start notifications
• Improved handling of processors that do not implement optional saving methods

Camera and preview lifecycle

Added

• Explicit handling for cameras that are starting, running, or stopping
• Support for cancelling preview while cameras are still opening
• Runtime reporting of actual camera frame rate, resolution, pixel format, and output format

Changed

• Stop Preview remains available during camera startup
• User-configured camera order is preserved in the tiled display while trigger startup ordering is handled separately
• Stable internal camera identities are now separated from human-readable display labels

Fixed

• Improved partial multi-camera startup when some cameras open successfully and others fail
• Prevented cameras from reporting a successful start after startup was cancelled
• Improved cleanup of partially initialized backends and worker threads
• Improved detection and rejection of duplicate camera configurations
• Improved recovery from camera disconnections, repeated empty frames, and startup failures
• Improved controller shutdown finalization

GUI and usability

Added

• Help → View documentation action for opening the DeepLabCut-Live-GUI documentation
• Detected camera resolution, frame rate, pixel format, and output format in the camera configuration dialog
• Improved recording path previews with session and run-folder placeholders
• Persistent preferences for recording names, processor selection, inference camera, and encoding options

Changed

• Improved camera discovery, configuration, reset, and preview workflows
• Improved camera and recorder status summaries
• Improved recording session naming and container-aware filename handling
• Improved control availability during camera startup, inference, recording, and shutdown

Fixed

• Improved asynchronous error reporting so background failures are visible in the GUI
• Improved cleanup when closing camera dialogs or the main application window
• Improved handling of unavailable cameras restored from saved configurations
• Improved configuration persistence and restoration across application sessions

Reliability and diagnostics

Added

• Timing diagnostics for camera workers, controller dispatch, inference, and recording
• Expanded automated coverage for camera backends, trigger modes, recording, timestamps, processor hooks, and GUI lifecycle behavior

Changed

• Improved logging for camera discovery, runtime information, recording state, and processor lifecycle events
• Expanded backend contract tests for camera reads, trigger waits, timestamp propagation, and cleanup

Fixed

• Improved handling of stale camera identities and GenTL producer paths
• Improved best-effort cleanup without hiding the original failure
• Improved consistency between camera backend results and downstream capture, inference, and recording code

What's Changed

• Update README install instructions for PyPI release by @C-Achard in #64
• Fix save_original bug in the processors by @arturoptophys in #75
• Add validation + compatibility tests for DeepLabCut-live by @deruyter92 in #56
• Several fixes to worker logic by @C-Achard in #57
• Upgrade & fix GenTL multi-camera mode by @C-Achard in #76
• Feature: triggered cameras, master and follower scheme by @C-Achard in #96
• Feature: encoding, UI and performance improvements by @C-Achard in #114
• 2.0 changelogs by @C-Achard in #116

Full Changelog: v2.0.0rc1...v2.0.0

Version 2.0 is a major rewrite of the application from scratch, delivering a redesigned GUI powered by the latest DLC-Live pose estimation workflows.

New in 2.0.0rc1

• New GUI rebuilt from the ground up using PySide6
• PyTorch support for DeepLabCut Live inference
• New camera backends for broader hardware support and more flexible acquisition setups

Highlights

Improved experience for live preview, inference, and recording
Better support for multi-camera experimental workflows
Modernized architecture to support future extensions and backend improvements

Notes

This is a major update and may differ from previous versions in setup, supported workflows, and backend behavior.
Please review the updated documentation for installation, camera configuration, and model compatibility.

Credits

• @arturoptophys PySide6 rewrite, new preview, controls & camera backends
• @C-Achard Production polish, UI/UX improvements, camera backends improvements, documentation
• @deruyter92 Code review, testing and feedback
• @MMathisLab Project lead

What's Changed

• Support Basler camera a2A2590-60umBAS (and other Basler cameras, untested) by @antortjim in #11
• Basler camera support by @MMathisLab in #12
• Relax python version upper bound by @jeylau in #20
• Upgrade Aravis backend features and testing by @C-Achard in #42
• Improving testing coverage and pre-commit checks before release by @C-Achard in #43
• Refactor camera config dialog and reduce code duplication by @C-Achard in #46
• Final UI tweaks by @C-Achard in #47
• Improve camera config loading UX and improve Basler backend by @C-Achard in #45
• Upgrade Gentl backend by @C-Achard in #44
• README update for PySide6 version by @C-Achard in #50
• Add standardized linting workflow and config by @C-Achard in #53
• Upgrade Gentl backend cti discovery by @C-Achard in #51
• Pre-release cleanup for PySide6 GUI by @C-Achard in #54
• Update dlclive requirement & add deploy workflow by @C-Achard in #48
• Deploy workflow candidate for 2.0.0rc1 by @C-Achard in #55
• Upgrade PySide6 multi-camera GUI by @C-Achard in #38
• Remove .coveragerc (duplicated in pyproject.toml) by @C-Achard in #61
• Trigger CI on main/master pushes and tags by @C-Achard in #62
• Add docs link to README by @C-Achard in #63

New Contributors

• @antortjim made their first contribution in #11
• @MMathisLab made their first contribution in #12
• @jeylau made their first contribution in #20
• @C-Achard made their first contribution in #42

Full Changelog: v1.0...v2.0.0rc1

The DeepLabCut-Live GUI is released! Simply pip install deeplabcut-live-gui