################################################################################
# SPDX-FileCopyrightText: Copyright (c) 2020-2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
################################################################################

Prequisites:
- DeepStreamSDK 6.1.1
- NVIDIA Triton Inference Server
- Python 3.8
- Gst-python
- NumPy

To set up Triton Inference Server:
For x86_64 and Jetson Docker:
  1. Use the provided docker container and follow directions for
     Triton Inference Server in the SDK README --
     be sure to prepare the detector models.
  2. Run the docker with this Python Bindings directory mapped
  3. Install required Python packages inside the container:
     $ apt update
     $ apt install python3-gi python3-dev python3-gst-1.0 python3-numpy -y

For Jetson without Docker:
  1. Install NumPy:
     $ apt update
     $ apt install python3-numpy
  2. Follow instructions in the DeepStream SDK README to set up
     Triton Inference Server:
     2.1 Compile and install the nvdsinfer_customparser
     2.2 Prepare at least the Triton detector models
  3. Add to LD_PRELOAD:
     /usr/lib/aarch64-linux-gnu/libgomp.so.1
     This is to work around the following problem with TLS usage limitation:
     https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91938
  4. Clear the GStreamer cache if pipeline creation fails:
     rm ~/.cache/gstreamer-1.0/*

To run the test app:
  $ python3 deepstream_ssd_parser.py <h264_elementary_stream>

This document shall describe the sample deepstream-ssd-parser application.

It is meant for simple demonstration of how to make a custom neural network
output parser and use it in the pipeline to extract meaningful insights
from a video stream.

This example:
- Uses SSD neural network running on Triton Inference Server
- Selects custom post-processing in the Triton Inference Server config file
- Parses the inference output into bounding boxes
- Performs post-processing on the generated boxes with NMS (Non-maximum Suppression)
- Adds detected objects into the pipeline metadata for downstream processing
- Encodes OSD output and saves to MP4 file. Note that there is no visual output on screen.

Known Issue:
1. On Jetson, if libgomp is not preloaded, this error may occur:
(python3:21041): GStreamer-WARNING **: 14:35:44.113: Failed to load plugin '/usr/lib/aarch64-linux-gnu/gstreamer-1.0/libgstlibav.so': /usr/lib/aarch64-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block
Unable to create Encoder

2. On Jetson Nano, ssd_inception_v2 is not expected to run with GPU instance.
Switch to CPU instance when running on Nano:
update config.pbtxt files in samples/trtis_modeo_repo:
# Switch to CPU instance for Nano since memory might not be enough for
# certain Models.

# Specify CPU instance.
instance_group {
  count: 1
  kind: KIND_CPU
}

# Specify GPU instance.
#instance_group {
#  kind: KIND_GPU
#  count: 1
#  gpus: 0
#}