Merge pull request #191804 from qm13/release1.1.1

v-regandowner · web-flow · commit 967e5b462c99 · 2022-03-15T18:02:27.000-04:00
Update documentation for release v1.1.1
diff --git a/articles/kinect-dk/body-sdk-download.md b/articles/kinect-dk/body-sdk-download.md
@@ -23,6 +23,7 @@ This document provides links to install each version of the Azure Kinect Body Tr
 
 Version       | Download
 --------------|----------
+1.1.1 | [msi](https://download.microsoft.com/download/9/d/b/9dbe0fbe-c9c3-4228-a64c-1e0a08736ec1/Azure%20Kinect%20Body%20Tracking%20SDK%201.1.1.msi) [nuget](https://www.nuget.org/packages/Microsoft.Azure.Kinect.BodyTracking/1.1.1)
 1.1.0 | [msi](https://www.microsoft.com/en-us/download/details.aspx?id=102901)
 1.0.1 | [msi](https://www.microsoft.com/en-us/download/details.aspx?id=100942) [nuget](https://www.nuget.org/packages/Microsoft.Azure.Kinect.BodyTracking/1.0.1)
 1.0.0 | [msi](https://www.microsoft.com/en-us/download/details.aspx?id=100848) [nuget](https://www.nuget.org/packages/Microsoft.Azure.Kinect.BodyTracking/1.0.0)
@@ -48,6 +49,21 @@ If the command succeeds, the SDK is ready for use.
 
 ## Change log
 
+### v1.1.1
+* [Feature] Added cmake support to all body tracking samples
+* [Feature] NuGet package returns. Developed new NuGet package that includes Microsoft developed body tracking dlls and headers, and ONNX runtime dependencies. The package no longer includes the NVIDIA CUDA and TRT dependencies. These continue to be included in the MSI package.
+* [Feature] Upgraded to ONNX Runtime v1.10. Recommended NVIDIA driver version is 472.12 (Game Ready) or 472.84 (Studio). There are OpenGL issues with later drivers.
+* [Bug Fix] CPU mode no longer requires NVIDIA CUDA dependencies [Link](https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1154)
+* [Bug Fix] Verified samples compile with Visual Studio 2022 and updated samples to use this release [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1250)
+* [Bug Fix] Added const qualifier to APIs [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1365)
+* [Bug Fix] Added check for nullptr handle in shutdown() [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1373)
+* [Bug Fix] Improved dependencies checks [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1510)
+* [Bug Fix] Updated REDIST.TXT file [Link](https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1541)
+* [Bug Fix] Improved DirectML performance [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1546)
+* [Bug Fix] Fixed exception declaration in frame::get_body() [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1573)
+* [Bug Fix] Fixed memory leak [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1576)
+* [Bug Fix] Updated dependencies list [Link] (https://github.com/microsoft/Azure-Kinect-Sensor-SDK/issues/1644)
+
 ### v1.1.0
 * [Feature] Add support for DirectML (Windows only) and TensorRT execution of pose estimation model. See FAQ on new execution environments.
 * [Feature] Add `model_path` to `k4abt_tracker_configuration_t` struct. Allows users to specify the pathname for pose estimation model. Defaults to `dnn_model_2_0_op11.onnx` standard pose estimation model located in the current directory.
diff --git a/articles/kinect-dk/body-sdk-setup.md b/articles/kinect-dk/body-sdk-setup.md
@@ -6,7 +6,7 @@ ms.author: quentinm
 ms.prod: kinect-dk
 ms.date: 06/26/2019
 ms.topic: quickstart
-keywords: kinect, azure, sensor, access, depth, sdk, body, tracking, joint, setup, cuda, nvidia
+keywords: kinect, azure, sensor, access, depth, sdk, body, tracking, joint, setup, onnx, directml, cuda, trt, nvidia
 ms.custom: mode-other
 #Customer intent: As an Azure Kinect DK developer, I want to set up Azure Kinect body tracking.
 ---
@@ -51,6 +51,45 @@ If everything is set up correctly, a window with a 3D point cloud and tracked bo
 
 ![Body Tracking 3D Viewer](./media/quickstarts/samples-simple3dviewer.png)
 
+## Specifying ONNX Runtime execution environment
+
+The Body Tracking SDK supports CPU, CUDA, DirectML (Windows only) and TensorRT execution environments to inference the pose estimation model. The `K4ABT_TRACKER_PROCESSING_MODE_GPU` defaults to CUDA execution on Linux and DirectML execution on Windows. Three additional modes have been added to select specific execution environments: `K4ABT_TRACKER_PROCESSING_MODE_GPU_CUDA`, `K4ABT_TRACKER_PROCESSING_MODE_GPU_DIRECTML`, and `K4ABT_TRACKER_PROCESSING_MODE_GPU_TENSORRT`.
+
+> [!NOTE]  
+> ONNX Runtime displays warnings for opcodes that are not accelerated. These may be safely ignored.
+
+ONNX Runtime includes environment variables to control TensorRT model caching. The recommended values are:
+- ORT_TENSORRT_ENGINE_CACHE_ENABLE=1 
+- ORT_TENSORRT_CACHE_PATH="pathname"
+
+The folder must be created prior to starting body tracking.
+
+> [!IMPORTANT]  
+> TensorRT pre-processes the model prior to inference resulting in extended start up times when compared to other execution environments. Engine caching limits this to first execution however it is experimental and is specific to the model, ONNX Runtime version, TensorRT version and GPU model.
+
+The TensorRT execution environment supports both FP32 (default) and FP16. FP16 trades ~2x performance increase for minimal accuracy decrease. To specify FP16:
+- ORT_TENSORRT_FP16_ENABLE=1
+
+## Required DLLs for ONNX Runtime execution environments
+
+|Mode      | ORT 1.10                       | CUDA 11.4.3          | CUDNN 8.2.2.26      | TensorRT 8.0.3.4 |
+|----------|--------------------------------|----------------------|---------------------|------------------|
+| CPU      | msvcp140                       | -                    | -                   | -                |
+|          | onnxruntime                    |                      |                     |                  |
+| CUDA     | msvcp140                       | cudart64_110         | cudnn64_8           | -                |
+|          | onnxruntime                    | cufft64_10           | cudnn_ops_infer64_8 |                  |
+|          | onnxruntime_providers_cuda     | cublas64_11          | cudnn_cnn_infer64_8 |                  |
+|          | onnxruntime_providers_shared   | cublasLt64_11        |                     |                  |
+| DirectML | msvcp140                       | -                    | -                   | -                |
+|          | onnxruntime                    |                      |                     |                  |
+|          | directml                       |                      |                     |                  |
+| TensorRT | msvcp140                       | cudart64_110         | -                   | nvinfer          |
+|          | onnxruntime                    | cufft64_10           |                     | nvinfer_plugin   |
+|          | onnxruntime_providers_cuda     | cublas64_11          |                     |                  |
+|          | onnxruntime_providers_shared   | cublasLt64_11        |                     |                  |
+|          | onnxruntime_providers_tensorrt | nvrtc64_112_0        |                     |                  |
+|          |                                | nvrtc-builtins64_114 |                     |                  |
+
 ## Examples
 
 You can find the examples about how to use the body tracking SDK [here](https://github.com/microsoft/Azure-Kinect-Samples/tree/master/body-tracking-samples).
diff --git a/articles/kinect-dk/reset-azure-kinect-dk.md b/articles/kinect-dk/reset-azure-kinect-dk.md
@@ -19,20 +19,24 @@ You may encounter a situation in which you have to reset your Azure Kinect DK ba
 
 1. Power off your Azure Kinect DK. To do this, remove the USB cable and power cable.
   ![A diagram that shows the location of the screw that covers the reset button.](media/reset-azure-kinect-dk-diagram.png)
-1. To find the reset button, remove the screw that's located in the tripod mount lock.
-1. Reconnect the power cable.
-1. Insert the tip of a straightened paperclip into the empty screw hole, in the tripod mount lock.
-1. Use the paperclip to gently press and hold the reset button.
-1. While you hold the reset button, reconnect the USB cable.
-1. After about 3 seconds, the power indicator light changes to amber. After the light changes, release the reset button.  
+2. To find the reset button, remove the screw that's located in the tripod mount lock.
+3. Reconnect the power cable.
+4. Insert the tip of a straightened paperclip into the empty screw hole, in the tripod mount lock.
+   >[!CAUTION]
+   >Never use a sharp ended tool like a pushpin to press the reset button. Instead use a flat ended tool like a paperclip to avoid damaging the reset button. 
+
+5. Use the paperclip to gently press and hold the reset button.
+6. While you hold the reset button, reconnect the USB cable.
+7. After about 3 seconds, the power indicator light changes to amber. After the light changes, release the reset button.  
    
    After you release the reset button, the power indicator light blinks white and amber while the device resets. 
-1. Wait for the power indicator light to become solid white.
-1. Replace the screw in the tripod mount lock, over the reset button.
-1. Use Azure Kinect Viewer to verify that the firmware was reset. To do this, launch the [Azure Kinect Viewer](azure-kinect-viewer.md), and then select **Device firmware version info** to see the firmware version that is installed on your Azure Kinect DK.
+8. Wait for the power indicator light to become solid white.
+9. Replace the screw in the tripod mount lock, over the reset button.
+10. Use Azure Kinect Viewer to verify that the firmware was reset. To do this, launch the [Azure Kinect Viewer](azure-kinect-viewer.md), and then select **Device firmware version info** to see the firmware version that is installed on your Azure Kinect DK.
 
 Always make sure that you have the latest firmware installed on the device. To get the latest firmware version, use the Azure Kinect Firmware Tool. For more information about how to check your firmware status, see [Check device firmware version](azure-kinect-firmware-tool.md#check-device-firmware-version).
 
+
 ## Related topics
 
 - [About Azure Kinect DK](about-azure-kinect-dk.md)
diff --git a/articles/kinect-dk/troubleshooting.md b/articles/kinect-dk/troubleshooting.md
@@ -185,48 +185,12 @@ The Sensor SDK C# documentation is located [here](https://microsoft.github.io/Az
 
 The Body Tracking SDK C# documentation is located [here](https://microsoft.github.io/Azure-Kinect-Body-Tracking/release/1.x.x/namespace_microsoft_1_1_azure_1_1_kinect_1_1_body_tracking.html).
 
-## Specifying ONNX Runtime execution environment
-
-The Body Tracking SDK supports CPU, CUDA, DirectML (Windows only) and TensorRT execution environments to inference the pose estimation model. The `K4ABT_TRACKER_PROCESSING_MODE_GPU` defaults to CUDA execution on Linux and DirectML execution on Windows. Three additional modes have been added to select specific execution environments: `K4ABT_TRACKER_PROCESSING_MODE_GPU_CUDA`, `K4ABT_TRACKER_PROCESSING_MODE_GPU_DIRECTML`, and `K4ABT_TRACKER_PROCESSING_MODE_GPU_TENSORRT`.
-
-> [!NOTE]  
-> ONNX Runtime displays warnings for opcodes that are not accelerated. These may be safely ignored.
-
-ONNX Runtime includes environment variables to control TensorRT model caching. The recommended values are:
-- ORT_TENSORRT_ENGINE_CACHE_ENABLE=1 
-- ORT_TENSORRT_CACHE_PATH="pathname"
-
-The folder must be created prior to starting body tracking.
-
-> [!IMPORTANT]  
-> TensorRT pre-processes the model prior to inference resulting in extended start up times when compared to other execution environments. Engine caching limits this to first execution however it is experimental and is specific to the model, ONNX Runtime version, TensorRT version and GPU model.
-
-The TensorRT execution environment supports both FP32 (default) and FP16. FP16 trades ~2x performance increase for minimal accuracy decrease. To specify FP16:
-- ORT_TENSORRT_FP16_ENABLE=1
-
-## Required DLLs for ONNX Runtime execution environments
-
-|Mode      | CUDA 11.1            | CUDNN 8.0.5          | TensorRT 7.2.1       |
-|----------|----------------------|----------------------|----------------------|
-| CPU      | cudart64_110         | cudnn64_8            | -                    |
-|          | cufft64_10           |                      |                      |
-|          | cublas64_11          |                      |                      |
-|          | cublasLt64_11        |                      |                      |
-| CUDA     | cudart64_110         | cudnn64_8            | -                    |
-|          | cufft64_10           | cudnn_ops_infer64_8  |                      |
-|          | cublas64_11          | cudnn_cnn_infer64_8  |                      |
-|          | cublasLt64_11        |                      |                      |
-| DirectML | cudart64_110         | cudnn64_8            | -                    |
-|          | cufft64_10           |                      |                      |
-|          | cublas64_11          |                      |                      |
-|          | cublasLt64_11        |                      |                      |
-| TensorRT | cudart64_110         | cudnn64_8            | nvinfer              |
-|          | cufft64_10           | cudnn_ops_infer64_8  | nvinfer_plugin       |
-|          | cublas64_11          | cudnn_cnn_infer64_8  | myelin64_1           |
-|          | cublasLt64_11        |                      |                      |
-|          | nvrtc64_111_0        |                      |                      |
-|          | nvrtc-builtins64_111 |                      |                      |
+## Changes to contents of Body Tracking packages
+
+Both the MSI and NuGet packages no longer include the Microsoft Visual C++ Redistributable Package files. Download the latest package [here](https://docs.microsoft.com/cpp/windows/latest-supported-vc-redist).
+
+The NuGet package is back however it no longer includes Microsoft DirectML, or NVIDIA CUDA and TensorRT files.
 
 ## Next steps
 
-[More support information](support.md)
+[More support information](support.md)
