blob: 9b9ea93730c83f45d86f330ccaa0466c8772b3c2 [file] [log] [blame]
Changes in 22.11
* Support for 22.11 NPU components and dependencies (core-driver, core-platform, Vela 3.6.0, CMSIS, TensorFlow Lite Micro)
* CMSIS-NN is added as a new dependency due to it moving out of the core CMSIS repository.
* Updates for Arm® Corstone™-310 platform.
* Formal support for CMake presets.
* Support for the new Arm Compiler 6.19
* Published CMSIS-Pack under `resources` and related updates.
* Bug fix for building with NPU disabled.
Changes in 22.08
* Support for 22.08 NPU component and dependencies (core-driver, core-platform, Vela 3.5.0, CMSIS, TensorFlow Lite Micro).
* Formal support for Arm® Corstone-310 Arm Virtual Hardware and FPGA implementations.
* Added CMake presets.
* Bug fix for Arm® Ethos-U65 NPU's default AXI burst length.
* Provision to allow platform-agnostic use case API to be packaged as CMSIS pack.
* Documentation updates.
Changes in 22.05
* Support for 22.05 NPU component and dependencies (core-driver, core-platform, Vela 3.4.0, CMSIS, TensorFlow Lite Micro).
* Restructuring of repository sources to allow generation of CMSIS packs.
* Added support for Arm Cortex-M85, Armclang 6.18 is required
* Reduce minimum requirements of Python to 3.7 to align with Vela 3.4.0.
* Increase minimum requirements of CMake to 3.21.0
* Documentation updates.
* Various bug fixes.
Changes in 22.02
* Support for 22.02 NPU component and dependencies (core-driver, core-platform, Vela 3.3.0, CMSIS, TensorFlow Lite Micro).
* Added Object Detection use case.
* Replace DSCNN with MicroNet for KWS and KWS_ASR use cases.
* Increase minimum requirements of CMake to 3.15.6, Armclang to 6.16 and Python to 3.8
* Initial restructuring of repository sources.
* Documentation updates.
* Various minor bug fixes.
Changes in 21.11
* Support for 21.11 NPU component and dependencies (core-software, core-driver, Vela 3.2.0, CMSIS, TensorFlow Lite Micro).
* Added dynamic load support for FVP for inference runner use-case.
* Added support for different memory modes: Shared_Sram, Dedicated_Sram and Sram_Only.
* Added Noise Reduction use case.
* Updated support for Arm GNU Embedded Toolchain 10.3-2021.07 and Arm Compiler 6.17.
* Documentation updates.
* Changes to support AN552 design - the new Arm® Corstone™-300 implementation.
Changes in 21.08
* Support for 21.05 NPU component (core-software, core-driver, Vela 3.0.0).
* TensorFlow submodule changed to (tested with TensorFlow Lite Micro commit hash: f510d38d0eaa3195ce3af66e3f32648740f08afb).
* Image classification model changed (from uint8 to int8 datatype).
* Documentation updates, added support to Corstone-300 + Arm® Ethos-U65.
* Various scripts improvements.
* Added Visual Wake Word use case.
Changes in 21.05
* Added script to download and optimize default models.
* Added script to run default build flow.
* Added a model for Anomaly Detection use case.
* Added support for build with Arm GNU Embedded Toolchain (10.2.1).
* Deprecated support for target subsystem SSE-200.
Changes in 21.03
* Simple platform support added.
* Model conditioning examples added.
* Documentation updated.
* Build changed to use sources of the dependency libraries.
* Tests for native platform added.
* Anomaly detection use-case added.
Changes in 20.11
* SSE-200 and SSE-300 system support was added.
* Support for simple fixed virtual platform for Arm® Ethos-U55 and Arm® Cortex-M55 removed.
* Build cmake parameters changed: TARGET_SUBSYSTEM was added, TARGET_PLATFORM accepted values were changed.
* Models with multiple output tensors support was added.
* Generic inference runner use-case added.
* ASR triggered by KWS added in the same use case (kws_asr). This also shows how to re-use tensor arena with two models using the same memory pool.
Changes in 20.09 release:
* Support for TensorFlow Lite Micro version > 2.3.0 (tested with TensorFlow Lite Micro 2.4.0 commit hash: 5bbb8a2bd1def6865b1510175a3da5fd12387e10)
* Added speech recognition use case example.
* Updated NPU Fastmodel version to r0p2-00eac0-rc4
Changes in 20.08 release:
* Added keyword spotting use case example.
* Added person detection use case example.
Known issues:
* telnet connection to FastModel environment may hang after some period of inactivity.
Changes in 20.05 release:
* FastModel environment was built with FastModel Tools v11.10.22.
* Mps3 FPGA build support was added.
* Configurable timing-adaptor.
* Added Active and Idle cycle counts for NPU and CPU profiling report.
* Source code structure and build scripts refactored to support multiple ML use-cases.
* Used EAC NPU software model and drivers.
* Windows support for build scripts.
Known issues:
* telnet connection to FastModel environment may hang after some period of inactivity.