Skip to content

Latest commit

 

History

History
200 lines (139 loc) · 9.24 KB

v2_0_0.md

File metadata and controls

200 lines (139 loc) · 9.24 KB

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

Full Changelog: https://github.com/DeepLabCut/DeepLabCut-live-GUI/compare/v2.0.0rc1...v2.0.0