release_notes.md

March 25, 2026 ยท View on GitHub

Release 11.2

Updated to match latest edgeai-tidl-tools release (This restructured version edgeai-tidl-tools has an easier to use interface and scripts)

Our recommended tool for model benchmark is now edgeai-tidlrunner instead of edgeai-benchmark. It supports a simple command line and config file based interfaces that are easier to use.

Release 11.1

Updated to match latest edgeai-tidl-tools release

Models that are not enabled by default in edgeai-benchmark (model_shortlist=100) have been removed from edgeai-modelzoo as well.

Example config files for models (_config.yaml file for each model) have been made more consistent - these now mostly only have parameters that are related to the model properties.

Release 11.0

New models in edgeai-modelzoo / edgeai-benchmark

We are in the process of adding support for several new models. Configs for models verified in this release are in this repository and the models are available in edgeai-modelzoo. The following new models have been verified:

Model nameModel TypeSource repository
rtmdet lite version (multiple flavours)Object Detectionedgeai-mmdetection
fastbev (without temporal)Multi-view 3DOD for ADASedgeai-mmdetection3d
bevformer_tinyMulti-view 3DOD for ADASedgeai-mmdetection3d

Note: The 3DOD models are trained with pandaset dataset (which is a Multi-view, Multi-modality ADAS / Automous Driving Dataset). edgeai-mmdetection3d and edgeai-benchmark now supports pandaset dataset. See more details of this dataset in edgeai-mmdetection3d.

Update on 2025 July 15

  • Accuracy fix for object detection models in edgeai-modelmaker and edgeai-mmdetection

Release 10.1

edgeai-benchmark core runtime apis

[2025-03-29 update] edgai-benchmark/edgeai_benchmark/core contains simplified wrapper apis over the core runtimes that are easy to use and understand. See edgeai-benchmark/run_tutorials_script_pc.sh and edgeai-benchmark/tutorials to understand the usage of these core apis.

New models in edgeai-modelzoo / edgeai-benchmark

We are in the process of adding support for several new models. Configs for models verified in this release are in this repository and the models are available in edgeai-modelzoo. The following new models have been verified:

Model nameModel TypeSource repository
swin_tiny_patch4_window7_224Image Clasifiation, Transformeredgeai-hf-transformers
deit_tiny_patch16_224Image Clasifiation, Transformeredgeai-hf-transformers
levit_128_224Image Clasifiation, Transformeredgeai-hf-transformers
segformer_b0Semantic Segmentation, Transformeredgeai-hf-transformers
YOLOv7 (mutliple flavours)Object Detectionedgeai-mmdetection
YOLOv9 (multiple flavours)Object Detectionedgeai-mmdetection
efficientdet_effb0_bifpn_liteObject Detectionedgeai-mmdetection
YOLOXPoseKeypoint detection / Human Poseedgeai-mmpose

Note: some of these models don't work natively in 10.1, but require a firmware update in the SDK and model compilation with firmware version set corresponding to that in tidl-tools. A separate script called run_benchmarks_firmware_update_pc.sh is provided in edgeai-benchmark to compile models with newer firmware.

New models in edgeai-modelmaker

Model nameModel TypeSource repository
YOLOv7Object Detectionedgeai-mmdetection
YOLOv9Object Detectionedgeai-mmdetection
YOLOXPoseKeypoint detection / Human Poseedgeai-mmpose

More details are in the Release Notes

Tech reports

New Tech report on 3D Object Detection - see the section on "Tech Reports"

Deprecations

edgeai-yolox repository is being deprecated - use edgeai-mmpose for Keypoint detection and edgeai-mmdetection for detection YOLOX models. The previous version of edgeai-yolox is still available in the previous branches of this repository or here

Release 10.0 (10_00)

  • Git branch: r10.0

Updates

Date: 18 October 2024

  • git tag: v10.0.2
  • tidl_tools update - bug fixes: edgeai-benchmark setup_pc.sh has been updated to install tidl_tools version 10_00_08_00. See release notes of edgeai-tidl-tools to understand the bug fixes that it has.

Date: September 2024

  • git tag: v10.0.1
  • Tech Reports added(Object detection, PyTorch FX based Quantization) - see this link
  • edgeai-hf-transformers (fork of https://github.com/huggingface/transformers) repository added, enabling training of Transformer based Image Classification, Segmentation and Object Detection Models.
  • Improvements for Model Surgery and Quantization in edgeai-modeloptimization
  • Several additional models added in edgeai-modelzoo and edgeai-benchmark
  • edgeai-mmpose(fork of open-mmlab/mmpose for efficient keypoint detection), edgeai-hf-transformers(fork of huggingface/transformers)
  • Tools: edgeai-tidl-tools version used: 10_00_06_00
  • Source code
  • Model Zoo
  • Model Zoo Release Manifest

Release 9.2 (09_02_00)

Release 9.1 (09_01_00)

Release 9.0 (09_00_00)

Release 8.6 (08_06_00)

Release 8.2 (08_02_00)

Additional software & models

Release 8.1 (08_01_00)

  • Date: December 2021

ModelZoo and compiled Model Artifacts

Other changes

  • This repository has been restructured. We now use the original mmdetection and added our changes on top of it as commits. This will help us to rebase from mmdetection as more and more features are added there. Please see more details in the usage.
  • We now have a tool that can automatically replace some of embedded un-friendly layers.

Additional models