Release Notes
Last updated
Last updated
November 4, 2024
Download Impulse experiments as CSV
November 4, 2024
Add anomaly head profiling to the memory performance
Implement object tracking/counting in EI SDK
Fixed undefined exception on reading 'inputBlocks'
October 31, 2024
Run batch sample operations on large datasets within a job
Add support for RPi bookworm & Microchip SAMA7 on Yocto
October 29, 2024
October 27, 2024
October 25, 2024
October 24, 2024
October 23, 2024
October 19, 2024
October 16, 2024
October 15, 2024
October 14, 2024
October 11, 2024
October 10, 2024
Check out the announcement blog!
October 8, 2024
October 7, 2024
October 4, 2024
Check out the announcement blog!
October 2, 2024
September 30, 2024
September 19, 2024
September 17, 2024
View the announcement blog!
September 17, 2024
September 13, 2024
September 12, 2024
September 5, 2024
August 29, 2024
August 29, 2024
August 28, 2024
August 28, 2024
August 27, 2024
August 26, 2024
August 22, 2024
August 21, 2024
August 20, 2024
August 15, 2024
August 15, 2024
Added the ability to create a new experiment when making changes to an existing impulse
Added flag that marks an impulse as stale following a change to the dataset
Added filtering to Experiments table
August 15, 2024
August 13, 2024
August 7, 2024
August 2, 2024
July 30, 2024
This Impulse design overhaul will enhance your model workflows and accelerate your project development cycles by allowing users to experiment with different impulses simultaneously, along with many other upgrades.
Check out the feature announcement blog!
July 24, 2024
July 19, 2024
View the documentation!
July 18, 2024
July 18, 2024
July 17, 2024
July 12, 2024
July 9, 2024
July 6, 2024
July 5, 2024
July 4, 2024
Check out the documentation!
July 4, 2024
July 2, 2024
July 1, 2024
July 1, 2024
June 28, 2024
June 28, 2024
June 26, 2024
June 26, 2024
June 21, 2024
View the public release announcement!
June 19, 2024
June 11, 2024
June 7, 2024
June 4, 2024
June 3, 2024
June 3, 2024
May 30, 2024
May 30, 2024
May 27, 2024
May 25, 2024
May 23, 2024
May 22, 2024
May 21, 2024
May 17, 2024
May 16, 2024
May 15, 2024
May 14, 2024
View the Getting Started documentation!
May 13, 2024
May 13, 2024
May 10, 2024
May 10, 2024
May 9, 2024
May 7, 2024
May 1, 2024
April 29, 2024
Various performance improvements and bug fixes
April 17, 2024
April 15, 2024
Various bug fixes
April 23, 2024
View the announcement blog!
April 9, 2024
April 2, 2024
March 27, 2024
March 18, 2024
March 14, 2024
March 14, 2024
View the Getting Started documentation!
March 12, 2024
March 8, 2024
March 8, 2024
Read the documentation!
February 29, 2024
February 27, 2024
View the announcement blog
February 26, 2024
February 17, 2024
February 16, 2024
February 12, 2024
February 12, 2024
February 9, 2024
February 8, 2024
February 6, 2024
February 5, 2024
February 2, 2024
January 20, 2024
January 18, 2024
January 18, 2024
January 17, 2024
January 17, 2024
January 13, 2024
January 10, 2024
January 10, 2024
January 9, 2024
January 9, 2024
January 8, 2024
January 5, 2024
January 4, 2024
January 3, 2024
Fix prediction positions in fit-long overlay comparison
Show experiment source (Manual vs from EON Tuner)
Fix EON Tuner for input type 'features'
Remove .gitmodules from Alif firmware
Change "tags" UI in impulse list to be more compact
Fix raw features order on live classify
Show ms in validation page if below one sec
Add `includes' to Arduino library properties
Fix the numbers by the icons in the EON tuner results for PatchCore
Data acquisition: small ID column changes
Ensure samples use DSP block axis order during feature generation
Implemented batch metadata updates in data acquisition
Implemented new EON tuner objective selection
Re-introduce tuner link to left navbar
Enable YOLOv5 for Himax WiseEye2
Add a helper function for users to resize matching Studio resize
Fix regression on YOLOV5 for DRP-AI
Make HR/HRV block available for all users
Add Int8 inference during training of FOMO-AD-Patchcore
Move all custom block configs to parameters.json
Adds more grayscale weights for MobileNetV2
Add multi-label support for SA and specgram autotuner
Support expanded view in live classification for visual AD result image
Allow keeping same order when uploading items in the UI
Release PatchCore for FOMO-AD for Visual Anomaly Detection learning block
Better BYOM navigation from dashboard
Show all available model types on model testing/live classification
Inform users when new impulse is not visible due to table filters
Notify users of CSV wizard config before uploading file
Fix live classification for projects with unused DSP blocks
Alif: add nproc info for MAC
Add SD card for BrickML
Add RAM/ROM estimates in the descriptions for visual AD backbones
Particle limit status readout during init
Ambiq Apollo4 integration
Support all resize methods in object detection overlay
Tabular data ingestion
Update to Openmv firmware and python scripts
Improvements to user experience with organization buckets
Add ability to add tags to impulse experiments
Add multilabel upload to Python SDK
Invalidate model testing summary when dataset changes
Add Syntiant DSP support to EON tuner
Add resize mode to exports and Linux runner
Added visual anomaly search space template to EON Tuner
Add support for multiple learn block in the EON Tuner
Update Vela compiler to 3.12.0
Update to sign-up flow
Show test set accuracy for regression and visual anomaly detection
Updated full tflite to 2.16.1 and added Mac M1 libraries
Add support for TAO models (i.e. SSD, RetinaNet, YOLOv3 and YOLOv4) for Ethos Linux inferencing engine
Added Raspberry Pi 5 as a new target
Alif new camera support
Add support for non-timeseries features
Impulse experiments improvements:
Make an explicit copy button for API/HMAC keys
Restore CMSIS NN optimizations for M0(+) MCUs for x2 INT8 speedup
Onnx2tf upgrade
Add scatter_nd, reduce_any, reduce_all ops
Fixed EON tuner trials failing when project contains only training dataset and no test dataset
Fix inference results on Himax WiseEye2 using Ethos
HD FOMO-AD
Impulse Experiments
Nicla OpenMV Runtime fix
Himax WiseEye2 (aka Grove Vision AI Module V2) integration
Microphone integration for Alif gen2
Allow custom learn blocks to specify output files
Python SDK - Add build() function to build impulse
Release EON tuner regression support
DSP Autotuner: padding, multi-label fixes + pass sample labels
Show label options in upload UI when 'unlabeled' selected
Make the Ethos U55-64 deployment public
Allow pro users to deploy classical ML
Alif gen2 IMU integration
Hide "items" for default dataset pipelines
Adds HRV and Motion Artifact Removal to HR DSP block
Retain filename when processing CSV data
Allow for JPEG images in SNAPSHOTSTREAM and SNAPSHOT mode
Release Synthetic data tab for pro users
Remove pro tier feature flag and user experiment
CSV Wizard: add support for one sensor value per row (same sample spanning multiple rows)
Update Arduino scripts
Linux CLI: Fix runner downloader for non-jetpack 6.0
Generate TVM library/EIM as standalone
Admin panel: jobs list on user, project and organization views
Fix exception on deployment for FOMO AD and EON RAM optimized mode
Show exact timestamp in grid view, synthetic data, model testing
Linux CLI: infer and automatically download for JetPack 6.0
Remove BETA badge and whitelisting for Microchip SAMA7G54 deploy option
Fix project search via WL admin API
Move perfcal graph back to scatter chart
Memryx aarch64 EIM support
Add Orin Inference and Build Container for JetPack 6.0
Think Silicon NEOX® AI SDK & GA100 RISC-V GPGPU
Added latency estimation for Microchip SAMA7G54
View auto-labeler segment improvements
Added Visual AD output to Arduino sketches
Fix wavelet visualization bug
Increase Jetson deployment memory
Add RZ/V2L TVM EIM deployment
Enable EON RAM Optimized mode on Arduino/Particle
EON Tuner: fix model testing, respect zero padding
Ensure tooltip targets are specified correctly
Model testing: Return 'float32' model variant when all blocks are anomaly
Fix Units of "By Frequency" slider from ms to Hz
Added yml to ethos pack deployment
Enable run static impulse in Sony firmware
Use aligned buffer for Nicla Vision framebuffer (ingestion fix)
For CMSIS pack also deploy a model.clayer.yml file
Anomaly block: min. confidence rating resets to 0 every time you train
Printf float for anomaly
Add Microchip SAMA7G54 deploy option to studio
Allow viewing where a segment came from in auto-segmenter
Launch of the "Professional Plan" pricing tier, designed to bridge the gap between the Community and Enterprise plans!
Akida Models part of EON Tuner Search Space
himax-flash-tool: Add support for Grove Vision AI Module V2
Fix 'this model won't run on MCU' for random forest binary classification
Ethos pack deployment
Fix to EON-compiled Ethos deployment
Properly sanitize labels when passing to Python/C++ code
Add performance calculation for HRV block
Fix feature explorer on restored projects
nRF7002DK Public Release
Use model threshold in mobile client instead of hardcoded values
More RAM for Seeed Sensecap
Add an optional mask to org header images
Improvements for when we copy data from DSP->Learn, when we re-split data
Various Python SDK fixes
Reduce resources for admin api/panel
Reduce mobile client memory consumed
Add sorting to the project list
Fix regression y_true values for model testing
Enable selecting model variant in live classification and model testing
Add 'save' button to Keras expert mode
BYOM - Move 'Test sample' up, so it's visible with many classes
Update Python SDK and Python API Bindings docs
Print current project in edge-impulse-uploader
IMU improvement for Syntiant NDP120
Allow toggling the square mask when training object detection models
Improved Live Classification and Performance Calibration performance
Fixed camera color issue for ESP32 Arduino sketches
Enable the toggling of the anomaly heatmap overlay in visual anomaly detection
Enabling multiple learning blocks deployment for all Arm Ethos-U accelerators
Visual Anomaly Detection (FOMO-AD)
Add websocket support to the Python SDK
YOLOv2 support in Akida inferencing engine
Multi-factor authentication
Make the 8 hour org job limit configurable in the UI
Add multilabel support to Python SDK
Improvements to resampling algorithm
Add EON Tuner support to Python SDK
Public transformation blocks
WASM SIMD deployment for NodeJS
Allow White Label admins to delete users
Jetson Orin
EON RAM optimized (patch-based inference) release. Minimize the RAM usage of the model to enable customers to execute more sophisticated and precise models with the same hardware resources.
Add support for Image scaling -128..127 for quantized image models
Make anomaly grid score responsive in mobile client
Show block name in training settings for custom blocks
Add display category for organization Transfer Learning blocks
Default datasets for organization transformations
Basler cameras support in Linux CLI
Classical ML blocks
Allow users to transfer expired trial projects to personal workspace
Allow guest users to access the Datasets view
Consistent color for bounding boxes
Adds back microphone support for MemryX
Add grid view to data acquisition
Add ability to list/unlist a public project
Use xterm to render job logs
RASyn board update
Speed up project restoration project
Add TAO integration (YOLOv3, YOLOv4, SSD and RetinaNet) in the embedded SDK
Improved classification view for object detection projects
Faster project versioning
Improve DSP parameters validation before feature generation
Allow filtering on ID in Data Acquisition
Add a 'Download source code' option to all custom blocks
Release Docker container Studio deployment option
Catch live classification with unsupported data
EON Compiler Subgraph Support
TAO SSD support
Python SDK - Update upload_pandas parameters, update readme and changelog
Add version information for EI-SDK pack
Launch VeLO for all enterprise customers
Add ID column to "Data acquisition"
Allow import/export CSV wizard config; and allow downloading original file used in CSV wizard
Added flatten DSP block support to EON tuner
Updated firmware for Arduino Nano 33
Add VeLO to EON Tuner and change name of config parameter
interval_ms should be a positive number, and should not be Infinity
Show EON Compiler toggle for GMM
EON Tuner fixes for custom ML blocks + autotuned blocks
Add OpenMV RT1062 to openmv lib deployment
Add more logging when sampling_freq missing
Fix for microphone ingestion for the Arduino Nano
Enabled camera sketch for Sony
Fix live classification for existing object detection models
Add CMSIS package deployment for Keil MDK
Add Studio version information to model_metadata.h
Downgrade arduino:mbed_nicla version to 3.4.1
Add 'Community' to 'Terms of Service' on signup pages
Generate and show logs for individual EON Tuner trials