Release notes version 08.02.00

New in this Release

  • Support for Sony’s IMX390 FPD link based sensors and Fusion1 board

  • Support for 8 channel capture of Sony’s IMX390 FPD link based sensors at 2MP @ 30fps using virtual channels

  • Support for Sony’s IMX219 Rpi camera with 10-bit capture support

  • Multi-instance Human pose estimation demo in edge_ai_apps Human Pose Estimation

  • Updated edgeai-gst-plugins with multi-channel support

  • Updated tiovxisp <https://github.com/TexasInstruments/edgeai-gst-plugins/wiki/tiovxisp> Gstreamer element with IMX390 support

  • Improved tiovxdlpreproc <https://github.com/TexasInstruments/edgeai-gst-plugins/wiki/tiovxdlpreproc> Gstreamer element with DMA optimizations to provide 2x performance lift

  • New tiovxdlcolorconvert Gstreamer element with DMA optimizations to provide 4x performance lift

  • Updated performance visualization tool with realtime junction temperature readings of CPU, GPU, C7x DSP and R5F cores

  • Validation of Intel-9260 WiFi card <https://www.amazon.in/Intel-Wireless-Ac-9260-2230-Gigabit/dp/B079QJQF4Y> in Access Point and Station mode

  • Enablement of Pulse Width Modulation (PWM) hardware accelerator

Fixed in this Release

  • ADASVISION-5149 - Hostname incorrect in SK shows j7-evm instead of tda4vm-sk

  • ADASVISION-5117 - USB 3.0 device is recognized as USB 2.1 sometimes

  • ADASVISION-4953 - Video looping does not work with v4l2 HW decoder plugin

  • ADASVISION-4943 - IMX219 based Rpi camera cannot capture data correctly for 10 bit format

  • ADASVISION-4809 - Python and C++ apps cannot measure pre and post processing time accurately

Known Issues

  • ADASVISION-5463 - ONR-OD-8000 model fails with IMX390 in CPP app due to 1200x1200 resolution

  • ADASVISION-5113 - GST Pipeline error: Error pulling tensor from GST Pipeline

  • ADASVISION-5070 - Edge AI demo output not resizing properly on 4K monitor

  • ADASVISION-4997 - Multi-input multi-inference demo freezes after long duration run

  • ADASVISION-4794 - Application takes 10-15 seconds to come up in some cases

  • ADASVISION-5454 - tiovxdlcolorcovert does not support UYVY

  • ADASVISION-5447 - Models failing due to tiovxdlcolorconvert limitation

  • ADASVISION-5421 - No metadata file was found for CPP apps with TVM models

  • ADASVISION-5420 - No valid frames found before end of stream

Model Zoo

About 83 models from 8.2 model zoo are regressed with multiple input sources such as camera and video files. A list of unsupported models are as below,

  • Height/2 is odd, tiovxdlpreproc block fetch logic fails

    • TFL-CL-0100-efficientNet-edgeTPU-m

    • TFL-CL-0140-efficientNet-lite4

    • TFL-CL-0170-efficientNet-lite1

    • TFL-CL-0190-efficientNet-edgeTPU-l

  • High latency causes DSS underflow

    • ONR-SS-8730-deeplabv3-mobv3-lite-large-cocoseg21-512x512

  • Odd resolution, tiovxdlcolorconvert fails

    • TVM-CL-3430-gluoncv-mxnet-xception

    • TFL-CL-0040-InceptionNetV3

  • Human pose model with unsupported post processing

    • ONR-KD-7000-human-pose-ae-mobv2-fpn-spp-udp-512x512

    • ONR-KD-7010-human-pose-ae-res50v2-fpn-spp-udp-512x512

    • ONR-KD-7020-human-pose-ae-mobv2-pan-spp-udp-512x512

    • ONR-KD-7030-human-pose-ae-res50v2-pan-spp-udp-512x512

Note

Please also refer to edgeai-gst-plugins issues for more details on bugs filed on custom GStreamer plugins

Software components

List of software components used in this version

Componenet

Version

PSDKLA (filesystem, uboot)

08.02.00.04

PSDKRA (firmware)

08.02.00.06

Docker

19.03.8-ce

Ubuntu (Docker)

20.04.LTS

Python

3.8

OpenCV

4.5.1

GStreamer

1.16.2

Cmake

3.16.3

Ninja

1.10.0

Meson

0.57.2

Jupyterlab

3.0.14

NeoAI - DLR

1.8.0

Tensorflow

TIDL_PSDK_8.2

TFLite-runtime

TIDL_PSDK_8.2

ONNX-runtime

TIDL_PSDK_8.2

PyYAML

5.4.1

TI Model Zoo

8.2.0

edgeai-gst-plugins

0.6.0

edgeai-tiovx-modules

8.2.0

edgeai-tidl-tools

8.2.0.5