Release Notes (2023)
Last updated
Last updated
December 23, 2023
BrickML SoM
View the !
December 21, 2023
Add run history to the EON Tuner, plus additional improvements
Add support for previewing Parquet files
December 21, 2023
Add ability to download data samples using the Python SDK
December 19, 2023
December 18, 2023
December 15, 2023
December 14, 2023
December 13, 2023
November 30, 2023
November 28, 2023
November 24, 2023
November 23, 2023
November 20, 2023
November 13, 2023
November 10, 2023
November 7, 2023
November 3, 2023
October 31, 2023
October 26, 2023
October 26, 2023
October 25, 2023
October 24, 2023
October 20, 2023
October 19, 2023
October 10, 2023
September 29, 2023
September 26, 2023
New deployment option: IAR Workbench
September 25, 2023
September 14, 2023
September 6, 2023
August 29, 2023
August 28, 2023
August 25, 2023
August 23, 2023
August 22, 2023
August 18, 2023
August 15, 2023
August 14, 2023
August 9, 2023
August 8, 2023
August 8, 2023
August 8, 2023
July 31, 2023
July 27, 2023
July 23, 2023
July 22, 2023
July 20, 2023
July 16, 2023
July 13, 2023
July 11, 2023
July 10, 2023
July 10, 2023
July 9, 2023
July 8, 2023
July 5, 2023
July 1, 2023
June 27, 2023
June 27, 2023
June 27, 2023
June 25, 2023
June 23, 2023
June 22, 2023
June 7, 2023
May 22, 2023
May 22, 2023
April 23, 2023
April 15, 2023
Single label classification formats: Plain CSV, Remo JSON
Object detection formats: Coco JSON, Plain CSV, Remo JSON, YOLO (YOLOv5, YOLOx), Open images CSV
April 6, 2023
April 4, 2023
EON Compiler v2
Native data onboarding for importing COCO, YOLO, Pascal VOC, and other standard image dataset types
March 31, 2023
March 13, 2023
March 6, 2023
February 28, 2023
February 24, 2023
February 22, 2023
February 15, 2023
February 10, 2023
February 8, 2023
January 20, 2023
January 17, 2023
January 16, 2023
January 10, 2023
January 3, 2023
Improved user experience for regression projects
Extended the EON Tuner to allow tuning object detection models. Read more on the updated .
Add Arduino sketches for Sony Spresense
(Enterprise only)
Eliminate page refresh when user adds themselves to a project within their Organization Project page
Enabled multi frequency for Thingy53 and Thingy91
Renesas EK-RA8D1
View the !
Allow White Label admins to generate compute usage reports
Improved logic to calculate compute time used by organizations
Improved functionality in our White Label Admin panel
Added support for ESP IDF (Espressif IoT Development Framework) 5.x
Enterprise users are now able to upload a random forest model (either LGBM or XGBoost)
Minor simplification to the DSP UI
Minor improvements to embedded SDK
Python SDK - Add public-facing helper functions to the Python SDK to get IDs and delete samples from a project.
Fixed a minor issue that at times was not allowing users to select items in data explorer
Extended the admin and white label admin APIs with project management endpoints. It also updates the admin panel with functionality to create free and enterprise projects and improves overall navigation
Accessibility and UI improvements in organization project view
Explain block versioning behavior in the UI
Update for Arduino Nano 33 Rev 2
Keep the same sample name after export/import
Add BrainChip YOLOv2 support
Particle Photon 2
Use the exact OpenAPI property names in serializing/deserializing objects
Configurable deadline for transform jobs
More control with linux classify camera examples
Asynchronously do RAM/ROM profiling in Keras blocks
Only include ei_vector.h if building for cpp
Split compute time metrics between GPU and CPU
Enable YOLOv5 inference for TensorRT Engine
Fix Akida deployment on AARCH64
Support time-series data with an interval of up to 24h
Sony multi-channel audio
Pass upload options to org transform blocks
TI AM68A
Fix that K-means clusters doesn't show in classification view
Anomaly detection GMM
Fix data explorer for multiple DSP blocks
Replace 'create new dataset' in 'add data item' with button
Small fixes to Akida custom profiling metrics
Data campaigns (Enterprise only)
Auto-labeling UI (Enterprise only)
Auto-zoom anomaly explorer
Show how long transforms / pipelines run in the UI
Ragged/numpy fix for BrainChip GPU FOMO training
Fix setting EI hosted portals as data source
Prevent double re-render of data breadcrumbs
FOMO GPU issue, do explicit numpy conversion before any iteration of ragged tensor
Fix CTA for EON Compiler on deployment page for BYOM
Allow openMV jobs container to run on Apple Silicon
Upgrade to TF2.11
Upgrade Silicon Labs targets to use latest SLC tooling
Store 'profile i8' setting in expert mode
BrickML
Himax AT Server Update (add AT+RUNIMPULSESTATIC)
Fix overflow of upload modal
Allow custom profiling metrics to show in Studio
Update Terms of Services view
TI AM68A
TI AM62A
Allow selecting 'unlabeled' on upload
Fix checkboxes for transform params in data sources
Show data with label "0.0" in data explorer
Add parameter UI for transformation blocks
Add HR block
CommonSense integration for Sony Spresense
Add ImageInput arg for deploy in Python SDK
Move input scaling into the Studio, and rework copyFeaturesFromDSP
GMM anomaly detection
Move "includedSamples" out of DSP metadata
Linux CLI: Fix detection of raspberry pi on bullseye
Allow white label admins to show/hide "Start with a tutorial" sections
Release data onboarding
Rework metrics widget for compute time and storage
Run process_output before profiling i8 model (yolo-models)
Profiling for non-visual GMM
Data onboarding fixes + UX improvements
Add data onboarding tools to CLI
Allow learn blocks to specify that they can only train on GPU
Fix EON Tuner custom block issues
Prevent page crash for EON Tuner projects with many classes
Add table with historic predictions to Linux/CLI tools
Fix IMU compilation for Nicla Voice
Enable building TAO containers in Kaniko
Edge Impulse CLI --download-data uses image input scaling from the block
Fix Nicla Voice inferencing with the right number of axes
Extend object detection classification results view
Allow selecting image input scaling
Add parameters.json to 'Edit block locally'
: Add microphone + IMU support
List uploaded pretrained models (and converted) under 'Downloads'
Added EON Tuner search space learn block attributes
Int8 quantization fix
Add options to neural compiler that are needed for YoloV5
PSoC 63: Add E-ink display handling
Firmware SDK update for Sony Spresense
Add BLE to PSoC 63 firmware and migrate to FreeRTOS
Allow uploading more image annotation formats:
Allow running DSP blocks with a features array
Today we are excited to launch a brand new Python SDK that connects Edge Impulse with the world’s leading ML tools for any edge hardware.
View the !
Build and deploy machine learning into edge products faster and more easily than ever before with these new features:
- Adding the ability to use pretrained models natively in Edge Impulse
to integrate Edge Impulse functionality into existing workflows
And , including and !
View the !
!
View the !
Data upload portals now work with mobile browsers
Developer profile - check it out on !
Added FOMO transformer support in SDK (quantized RSQRT ops)
View the !
FOMO support added for
Added support for transformer models in Edge Impulse (this adds support for SQUARED_DIFFERENCE, BATCH_MATMUL and EXPAND_DIMS)
Added support for YOLOv7 final layer (and YOLOv7 pre-trained models)
Added a library deployment option for
added ResNet support
View the
Added YOLOX library generation + EIM deployment options for
Allow DSP blocks to output TFLite files
Added EON tuner search space parameters
Add Arduino sketches for RP2040 (fusion and microphone), also compatible with Arduino RP2040 Connect sensors
View the !
Migrate Infineon PSoC firmware to Infineon SDK 3.0
View the !
View the !