Release Notes

V2.3.0 Change Logs (11/10/2021)

  • General

    • Released new firmware for EVK2, upgrade is required for compatibility with this MV2.3.0 release

    • Upgraded required pybind11 version from 2.4.3 to 2.6.0 when building from sources

    • Reorganized the test dataset archive to enable partial downloads

    • Reviewed the Ubuntu packages architecture for Essentials. The install command is now sudo apt -y install metavision-essentials, which includes the Python packages for the default Python version of your system. For other Python versions, you can also specific packages, e.g. package metavision-essentials-python3.7 for Python 3.7.

  • OpenEB

    • New features

      • Added Github Action CI pipeline on OpenEB repository (https://github.com/prophesee-ai/openeb)

      • Introduced new RAW plugins, adding capability to use RAW recording as input to processing pipelines

      • Introduced new RAW recordings seek capability, in beta version

      • Introduced new adaptive-rate event-stream slicing adapting event iteration to scene dynamics

      • Added Python sample demonstrating usage of various event-based formats (CSV, ROSbag, AEDAT) as input to Python pipelines

    • Updates

      • Added access to the ERC API from the Camera class

      • Moved definition of EventBbox from ML module to OpenEB and introduced EventBboxNpyReader class to handle various input formats

      • Added possibility to change default color palette in Python binding of PeriodicFrameGenerationAlgorithm

      • Renamed Gen41 FO bias from bias_fo_p to bias_fo

    • Bug fixes

      • Metavision::Window does not crash anymore when requesting window sizes larger than the screen resolution

  • Metavision

    • New features

      • Introduced Metavision Studio, a user-friendly viewer for live cameras and recordings

      • CV Module

        • Introduced new C++ and Python samples for data-rate evaluation

      • ML Module

        • Introduced new toolbox including training, inference and metrics for supervised classification

        • Introduced new model and pipeline demonstrating rock/paper/scissor gesture recognition

      • Calibration module

    • Changes

      • Machine Learning

        • Added possibility in the detection & tracking pipeline in ML module to update box tracklet coordinates after each inference

      • Analytics

        • Optimized runtime of PSM algorithm in Analytics module

    • Bug fixes

      • The Python bindings of the AntiFlickerAlgorithm can now be used correctly

  • Prophesee Devices

    • New features

      • Added support for trigger/sync mode, temperature and illumination to EVK3 Gen41

    • Changes

    • Bug fixes

  • Documentation:

    • Enhancement of installation and getting started pages

    • Enhancement of Python and C++ guides for SDK modules

    • Clarification of biases and triggers pages

V2.2.2 Change Logs (26/07/2021)

  • Cameras supported:

    • New plugin for Prophesee EVK3 Gen41 camera with limited set of features (streaming, biases, ROI). This plugin does not support Monitoring events for Illumination and Temperature. External Trigger mechanism (Master/Slave mode, Trigger In, Trigger Out) is not yet supported as well.

  • Bug fixes:

    • Machine Learning

      • Python package metavision_ml now deployed by Windows Essentials installer

      • The eval_coco_kpi sample does not fail anymore when no box is evaluated

    • Analytics

      • Vibration estimation is now using correct color map

      • Python Sample metavision_jet_monitoring_calibration now imports all the necessary modules

    • Miscellaneous

      • Enhancement of XYT Designer sample to show frame view next to events view

      • The install step after compilation now works without error on Python packages deployment path

  • Documentation:

    • Addition of an USB extension cable reference in the FAQ

    • Additional information in the FAQ about IMU deprecation

V2.2.1 Change Logs (11/06/2021)

  • Bug fixes:

    • Prophesee Camera plugins

      • The ERC facility setter functions now saturates to a device-dependent maximum value

      • The AFK facility setter functions now preserves the anterior enable/disable status

      • The stop parameter in the set_frequency AFK facility function is now correctly taken into account

      • The initial event burst occurring in some situations with Gen3.1 devices is now fixed

    • Machine Learning

      • The Optical Flow model now directly predicts in absolute pixel values

      • The detection models have been renamed to clarify their specificities

      • The HDF5Iterator now correctly stops at the end of the input file and correctly supports the start_ts argument

      • The train_detection sample now loads the right classes in finetuning mode and correctly supports the just_val flag

      • The export_detector sample can now be run from the build directory without installing first

      • The viz_event_simulator sample now correctly supports implicit input height & width

      • The colormaps used for visual flow visualizations are now consistent

    • Miscellaneous

      • The default installation path for Windows when building from source code with the deployment step is now C:\Program Files\Prophesee

  • Documentation:

V2.2.0 Change Logs (29/03/2021)

V2.1.0 Change Logs (22/09/2020)

  • Platforms supported:

    • Ubuntu 16.04 is not supported anymore

    • Python 3 support in Metavision Designer (instead of Python 2)

  • New features:

    • New Metavision SDK samples & applications:

    • New Metavision Designer samples:

      • Analytics samples: metavision_counting.py, metavision_generic_tracking.py

      • Metavision HAL (Hardware Abstraction Layer) with documented API, Prophesee camera plugins and code samples

      • Metavision Player (to visualize and record data streamed by event-based vision system)

    • Documentation online (no more delivered along with the software)

  • Deprecated features & software:

    • EM events

    • IMU events

    • Prophesee Player (replaced by new Metavision Player software)

    • SDK samples using the engine “_engine_sample”

  • Changes in existing features:

    • Multiple changes in the SDK and Designer API

      • SDK API: new modules organization in base, core, driver, cv

      • Designer API: new modules organization in base, core, engine, cv

    • SDK samples are not delivered as precompiled binaries (now, need to compile them)

    • SDK and Designer samples renaming

    • Simplifying datasets delivery (now, they can be downloaded from the public documentation)

    • ObjectPool moved from Metavision SDK Core module to Metavision SDK Base module. Use #include “metavision/sdk/base/utils/object_pool.h instead.