Feature/triton jetson support (#23)

* Testing Jetson support in gst-triton-parallel-tracking-v2.py

* Added queues

* triton parallel tracking v1 now uses nvmultiurisrcbin, gst-tracking-v2 uses nvurisrcbin

* Fix for writing out the file

* Added examples using gst-launch-1.0

* Run pre-commit

* Examples & improvements

.gitignore

@@ -134,6 +134,9 @@ dmypy.json
 # mp4 video files
 *.mp4
 
+# mkv video files
+*.mkv
+
 # Gaphviz dot files
 *.dot
 
@@ -149,3 +152,6 @@ dmypy.json
 
 # VS code
 .vscode/
+
+# TensorRT Engine files
+*.engine

deepstream-examples/README.md

@@ -15,14 +15,14 @@ List of examples:
 
 * [deepstream-tracking](deepstream-tracking/README.md)
   * 4-class object detector with tracking
-  * Tested with deepstream 6.1
+  * Tested with deepstream 6.3
 * [deepstream-tracking-parallel](deepstream-tracking-parallel/README.md)
   * 4-class object detector with tracking
   * Splits the input stream into two and runs two pipelines on the split streams
   * Tested with deepstream 6.1
 * [deepstream-triton-tracking](deepstream-triton-tracking/README.md)
   * 4-class object detector with tracking, uses local version of the Triton Inference Server for inference
-  * Tested with deepstream 6.1
+  * Tested with deepstream 6.3
 * [deepstream-retinaface](deepstream-retinaface/README.md)
   * RetinaFace bbox- and landmark detector
   * Uses a custom parser called [NvDsInferParseCustomRetinaface](src/retinaface_parser/nvdsparse_retinaface.cpp)
@@ -172,7 +172,7 @@ After this you can create the docker image used in the examples.
 
 ```bash
 cd gstreamer-examples/docker
-docker build -t nvidia-deepstream-samples -f ./Dockerfile-deepstream .
+docker build -t deepstream-6.3 -f ./Dockerfile-deepstream-6.3-triton-devel .
 ```
 
 ## 4.2 Test the Docker Image
@@ -202,7 +202,7 @@ docker run -i -t --rm \
   -e DISPLAY=$DISPLAY \
   -e XAUTHORITY=$XAUTHORITY \
   -e NVIDIA_DRIVER_CAPABILITIES=all \
-  --gpus all nvidia-deepstream-samples bash
+  --gpus all deepstream-6.3 bash
 ```
 
 Then execute the following inside the container:
@@ -252,9 +252,9 @@ docker run -i -t --rm \
   -e DISPLAY=$DISPLAY \
   -e XAUTHORITY=$XAUTHORITY \
   -e NVIDIA_DRIVER_CAPABILITIES=all \
-  --gpus all nvidia-deepstream-samples bash
+  --gpus all deepstream-6.3 bash
 cd /home/gstreamer-examples/deepstream-examples/deepstream-tracking
-python3 gst-tracking.py -i /opt/nvidia/deepstream/deepstream-6.1/samples/streams/sample_1080p_h264.mp4
+python3 gst-tracking.py -i /opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4
 ```
 
 When starting the Docker container with the above command, the switch `-v $(pwd):/home/gstreamer-examples` maps the local directory `$(pwd)`
@@ -302,7 +302,6 @@ pip3 install pyds-1.1.4-py3-none-linux_x86_64.whl
 
 Replace `pyds-1.1.4-py3-none-linux_x86_64.whl` with the version that you downloaded.
 
-
 ## 5.3 Install Triton Inference Server
 
 Before executing those examples that use Triton, you first need to install it locally. First install the following package(s):
@@ -372,7 +371,6 @@ cd /opt/nvidia/deepstream/deepstream/samples
 ./prepare_ds_triton_model_repo.sh
 ```
 
-
 ## 5.6 Testing Triton Installation
 
 Test that the `nvinferenceserver` plugin can be found

deepstream-examples/deepstream-tracking-parallel/README.md

@@ -1,4 +1,4 @@
-# Deepstream Tracking
+# 1 Deepstream Parallel Tracking
 
 This example shows to split an input stream into two, using a tee-element, so that two different image processing pipelines can process the same stream.
 This example processes the split streams using the same inference elements, but they can be different for each stream. It appears that you need to add an
@@ -9,15 +9,15 @@ nvstreammux-element into both of the processing streams, after the tee-element,
 * PGIE_CLASS_ID_PERSON = 2
 * PGIE_CLASS_ID_ROADSIGN = 3
 
-# Pipeline
+# 2 Pipeline
 
 Pipeline description.
 
 ![Image of the pipeline](./gst-tracking-parallel.pdf)
 
-# Processing Pipeline Configurations
+# 3 Processing Pipeline Configurations
 
-## Pipeline 1
+## 3.1 Pipeline 1
 
 Configuration files for the inference- and tracker elements:
 
@@ -32,7 +32,7 @@ Configuration files for the inference- and tracker elements:
 * Tracker
   * Configuration file: [tracker_config_1.txt](tracker_config_1.txt)
 
-## Pipeline 2
+## 3.2 Pipeline 2
 
 Configuration files for the inference- and tracker elements:
 
@@ -47,7 +47,7 @@ Configuration files for the inference- and tracker elements:
 * Tracker
   * Configuration file: [tracker_config_2.txt](tracker_config_2.txt)
 
-## Requirements
+## 3.3 Requirements
 
 * DeepStreamSDK 6.1.1
 * Python 3.8
@@ -57,7 +57,7 @@ Configuration files for the inference- and tracker elements:
 * gstreamer1.0-plugins-bad
 * gstreamer1.0-plugins-ugly
 
-## How to Run the Example
+## 3.4 How to Run the Example
 
 In order to get help regarding input parameters, execute the following:
 
@@ -84,3 +84,75 @@ The dot file can be converted into pdf as follows:
 dot -Tpdf <NAME-OF-THE-DOT-FILE> -o output.pdf
 ```
 
+# 4 Test Pipelines
+
+Following are test pipelines that can be launched with `gst-launch-1.0`. Requirements for running the pipelines:
+
+* Deepstream 6.3
+
+## 4.1 Processing Several Streams Using a Single Pipeline
+
+Figure 1. shows the pipeline. We connect several streams to a single processing pipeline. I have omitted some details
+from the pipeline so that it fits better on the screen.
+
+<figure align="center">
+    <img src="./figures/multi_input_pipeline.png" width="900">
+    <figcaption>Figure 1. Parallel processing of several streams with a single pipeline.</figcaption>
+</figure>
+
+### 4.1.1 Processing Pipeline with 4 Input Streams with Video- and Filesinks
+
+```bash
+gst-launch-1.0 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_0 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_1 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_2 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_3 \
+nvstreammux name=m width=1280 height=720 batch-size=4 ! nvinfer config-file-path=dstest2_pgie_config.txt batch-size=4 \
+model-engine-file=/opt/nvidia/deepstream/deepstream/samples/models/Primary_Detector/resnet10.caffemodel_b4_gpu0_int8.engine \
+! queue ! \
+nvtracker tracker-width=640 tracker-height=480 ll-config-file=config_tracker_NvDCF_perf_uniqueid.yml \
+ll-lib-file=/opt/nvidia/deepstream/deepstream/lib/libnvds_nvmultiobjecttracker.so ! queue ! \
+nvmultistreamtiler rows=2 columns=2 width=1280 height=720 ! queue ! \
+nvdsosd ! tee name=t \
+t. ! queue ! nvvideoconvert ! 'video/x-raw(memory:NVMM), format=NV12' ! nvv4l2h264enc profile=High bitrate=10000000 ! h264parse ! matroskamux ! \
+filesink location=4_stream_output.mkv \
+t. ! queue ! nvvideoconvert ! nveglglessink
+```
+
+### 4.1.2 Processing Pipeline with 20 Input Streams with Video- and Filesinks
+
+```bash
+gst-launch-1.0 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_0 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_1 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_2 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_3 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_4 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_5 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_6 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_7 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_8 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_9 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_10 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_11 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_12 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_13 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_14 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_15 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_16 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_17 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h264.mp4 ! queue ! m.sink_18 \
+nvurisrcbin uri=file:///opt/nvidia/deepstream/deepstream/samples/streams/sample_1080p_h265.mp4 ! queue ! m.sink_19 \
+nvstreammux name=m width=1280 height=720 batch-size=20 ! nvinfer config-file-path=dstest2_pgie_config.txt \
+batch-size=30 \
+model-engine-file=/opt/nvidia/deepstream/deepstream/samples/models/Primary_Detector/resnet10.caffemodel_b30_gpu0_int8.engine ! queue ! \
+nvtracker tracker-width=640 tracker-height=480 ll-config-file=config_tracker_NvDCF_perf_uniqueid.yml \
+ll-lib-file=/opt/nvidia/deepstream/deepstream/lib/libnvds_nvmultiobjecttracker.so ! queue ! \
+nvmultistreamtiler rows=5 columns=4 width=1280 height=720 ! queue ! \
+nvdsosd ! tee name=t \
+t. ! queue ! nvvideoconvert ! 'video/x-raw(memory:NVMM), format=NV12' ! nvv4l2h264enc profile=High bitrate=10000000 ! h264parse ! matroskamux ! \
+filesink location=20_stream_output.mkv \
+t. ! queue ! nvvideoconvert ! nveglglessink
+```
+