Jacinto 7 TIDL Release Notes

Version: 08.00.00, Date: July 31, 2021

Contents

  1. Introduction
  2. Licensing
  3. What's New
  4. Documentation
  5. Upgrade and Compatibility Information
  6. Device Support
  7. Validation Information
  8. Fixed Issues
  9. Known Issues
  10. Change Request
  11. Technical Support
  12. Package Versioning

Introduction

The package consists of Texas Instrument's Deep learning solution on Jacinto 7 product family

Licensing

The licensing information of this product, as well as any third-party components included which are made available under a number of other open-source licenses are enumerated as part of the manifest. Note : Below Links would work only after installing the package

Documentation

1 User's Guide [HTML]
2 Software Manifest [HTML]

What's New

Here are a few of the new features supported in this release:

  1. Improved Model zoo for camera applications
  2. Performance improvement of Object Detection Models for ONNX Runtime
  3. Performance improvement of data transfer and format conversion routines between ARM and DSP for TFLite and ONNX Runtime
  4. Optimal support of 5x5 depthwise convolution with stride 2
  5. Support of batch processing
  6. Improved Jupyter notebooks for better debug capabilities
  7. Improved graph visualization for complete model with annotation of target (C7x or ARM) for each sub graph

Upgrade and Compatibility Information

NA

Device Support

Target SoC Target Plaform Build environment (OS) Target environment (OS)
J721E EVM x86_64 Linux PC x86_64 (Linux) NA
J721E EVM x86_64 Windows PC x86_64 Windows NA

Note: Support of x86_64 Windows would be removed from next release onwards and only x86_64 Linux will be maintained

Validation Information

This release was built and validated using the following tools:

Dependencies (included in Processor SDK RTOS Automotive):

Dependencies (NOT included in Processor SDK RTOS Automotive):

Refer User's Guide for instructions to install and setup above dependencies.

Fixed Issues

ID Description Module Affected Versions Affected Platforms
TIDL-705 Max Pooling Output is not correct with signed input COMPUTE 01.00.01 J721E
TIDL-1157 Object Detection – YOLOv3 Significant Accuracy drop in float mode COMPUTE 01.03.00 J721E
TIDL-1182 Verbose info while giving segmentation fault TOOL 01.02.00 J721E
TIDL-1273 OpSet11 model generates many sub-graphs. TOOL 01.03.00 J721E
TIDL-1333 "ONNX Opset 11 with ""SIZES"" field as input tensor is not working" TOOL 01.04.00 J721E
TIDL-1338 TDA4VM: GlobalAvgPool layer is returning error during inference COMPUTE 01.04.00 J721E
TIDL-1342 3x3 Avg pooling will get stuck on EVM with small height COMPUTE 01.04.00 J721E
TIDL-1348 During Import of model Error Code = is reported TOOL 01.04.00 J721E
TIDL-1362 Update import documentation to remove network compiler estimate file Unknown 02.00.00 J721E
TIDL-1363 Documentation : Wrong unit is used for reporting performance in trace logs Common 02.00.00 J721E
TIDL-1367 TIDL Resize Layer output mismatch on C7x COMPUTE 02.00.00 J721E
TIDL-1368 Mixed Precision not working for Sigmoid COMPUTE 02.00.00 J721E
TIDL-1370 Mixed Precision flow may cause exception for BatchNorm/Eltwise layer on EVM COMPUTE 02.00.00 J721E
TIDL-1371 Undefined behavior for network with property as mentioned in description COMPUTE 02.00.00 J721E
TIDL-1375 Functional Mismatch for 5x5 stride-by-2 Valid Convolution COMPUTE 02.00.00 J721E
TIDL-1377 Tflite runtime : Fully connected layer with more than one variable inputs results in segmentation fault TOOL 02.00.00 J721E
TIDL-1379, TIDL-1474 DOCUMENTATION Gaps :
  • Incorrect Protobuf version in user guide
  • Improve documentation for padding and input normalisation(use of foldPreBnConv2D = 2 and inDataPadInTIDL=1)
  • Missing dependencies for open source run time import build in user guide
  • Meta arch type in user guide shows 0/1 whereas supported types are upto 5 currently
  • In TIDL supported layers list, need to mention argMax needs keepDims = 1 in addition to axis == 1
  • Missing document about shape inference to be must for ONNX runtime
  • Missing documentation on how to read final output of TIDL
  • Incorrect mention about CCS based standalone mode availability
  • PDK windows build documentation update
    		•  Common 02.00.00 J721E
    TIDL-1389 Custom layer : Constraints are missing from documentation COMPUTE 02.00.00 J721E
    TIDL-1390 Default value for bias calibration frames with inFileFormat = 1 is not correct TOOL 02.00.00 J721E
    TIDL-1394 Onnx runtime : Sigmoid and argmax allowlisting OSRT 02.00.00 J721E
    TIDL-1413 Accuracy Improvement: Output of Relu6 activation is not clipped to 6 COMPUTE, TOOL 02.00.00 J721E
    TIDL-1435 Object detection segmentation fault happens with top_k is set to 5000 COMPUTE 02.00.00 J721E
    TIDL-1468 Concatenate output error (tidl_j7_01_02_00_09) COMPUTE 01.04.00 J721E
    TIDL-1470 Run time error (core dump or segmentation fault) in graph compiler during import of network TOOL 02.00.00 J721E
    TIDL-1472 Segmentation models not running for 8 bit inference on EVM COMPUTE 02.00.00 J721E
    TIDL-1517 Tflite runtime : resize_tensor_input API not validated TOOL 02.00.00 J721E
    TIDL-1518 Mixed Precision : PC output not matching with EVM output for non-convolution layers COMPUTE 02.00.00 J721E
    TIDL-1519 Incorrect documentation to mention as pooling being merged with convolution Common 02.00.00 J721E
    TIDL-1520 ci and ref output mismatch for yolov5 model COMPUTE 02.00.00 J721E
    TIDL-1523 Error in TF import for Concat layer when axis is height or width COMPUTE 02.00.00 J721E
    TIDL-1529 Onnx runtime : Multiplication operator to be allowed only for elementwise or const multiplication OSRT 02.00.00 J721E
    TIDL-1531 TI Edge cloud gives erroneous results for classification TOOL 02.00.00 J721E
    TIDL-1534 Error in TfLite import for Inner product layer TOOL 02.00.00 J721E
    TIDL-1541 Onnx runtime : Running out of memory on EVM on running several models consecutively OSRT 02.00.00 J721E
    TIDL-1543 TI edgeai Cloud: Error report for Custom model compilation TOOL 02.00.00 J721E
    TIDL-1547 TIDL Sigmoid layer failures after updating IT quantization flow TOOL 02.00.00 J721E
    TIDL-1555 Conv2D behaves differently on PC and TDA4 with mixed precision TOOL 02.00.00 J721E

    Known Issues

    ID Description Module Reported in Release Affected Platforms Occurrence Workaround in this release
    TIDL-864 Concat trace dump mismatch between REF & CN/CI COMPUTE 01.02.00 J721E Rare NA
    TIDL-1197 Model execution order for a branch is not optimal COMPUTE 01.02.00 J721E Rare No fucntional issue, Only Performance drop would be obsrved
    TIDL-1269 TF maxpool avgpool functionality mismatch for k=3,s=2 COMPUTE 01.03.00 J721E Rare NA
    TIDL-1350 ShuffleNetV2 ONNX-RT accuracy is 2% below compared to TVM-RT COMPUTE 02.00.00 J721E Rare NA
    TIDL-1380 3D conv layer not marked as unsupported in allowlisting in open source run time OSRT 02.00.00 J721E Rare NA
    TIDL-1530 Processing on EVM gets stuck in case of "GEMM" layer in onnx model OSRT 02.00.00 J721E Rare set reservedCtrl=1 of TIDL_CreateParams
    TIDL-1546 Functional incorrect output with 3x3 stride 2 single channel convolution layer OSRT 02.00.00 J721E Rare NA

    For latest status of the issues and any new issues found post this release, please refer TI External software Incident Report. Issues on this product can be filtered with search query as project = EXT_EP AND issuetype in (Bug, Enhancement) AND Product ~ "TI Deep Learning Library"

    Change Request

    ID Description Original Fix Version New Fix Version
    JACINTOREQ-1537 Power Benchmarking in TIDL Cloud Evaluation Tool 08.00.00 Yet to decide

    Technical Support

    For technical support, please post your questions on TI E2E Forum for Automotive ADAS SoCs.

    For additional assistance, contact local TI Field Application Engineer

    Package Versioning

    Each package version is composed of 4 period-delimited numbers - represented here by the letters M, m, p and b [M.m.p.b]. The table below provides a descriptive reference regarding package version numbering.

    Digit Meaning Description
    1 (M=Major) Major revision Incremented when the new version is substantially different from the previous For example, a new module added or an existing module's algorithm significantly altered.
    2 (m=minor) Minor revision Incremented when the new version has changed but not in a major way. For example, some minor changes in the API or feature set.
    3 (p=patch) Patch number Incremented for all other source code changes. This include any packaging support code.
    4 (b=build) Build number Incremented for each release delivery to CM. Reset for any change to M, m or p

    © Copyright 1995-2021, Texas Instruments Incorporated. All rights reserved.
    Trademarks | Privacy policy | Terms of use | Terms of sale