README.md

################################################################################
#
# Copyright (c) 2020, NVIDIA CORPORATION. All rights reserved.
#
# 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.
#
################################################################################

YOLOv4 and DeepStream

This sample shows how to integrate YOLOv4 with customized output layer parsing for detected objects with DeepStreamSDK.

1. Sample contents:

• deepstream_app_config_yolov4.txt: DeepStream reference app configuration file for using YOLOv4 model as the primary detector.
• config_infer_primary_yolov4.txt: Configuration file for the GStreamer nvinfer plugin for the Yolo detector model.
• nvdsinfer_custom_impl_Yolo/nvdsparsebbox_Yolo.cpp: Output layer parsing function for detected objects for the Yolo model.

2. Pre-requisites:

2.1 Please make sure DeepStream 5.0 is properly installed

2.2 Generate YOLOv4 TensorRT Engine

• Go to this pytorch repository https://github.com/Tianxiaomo/pytorch-YOLOv4 where you can convert YOLOv4 Pytorch model into ONNX and TensorRT engine

• Other famous YOLOv4 pytorch repositories as references:

  • https://github.com/WongKinYiu/PyTorch_YOLOv4
  • https://github.com/bubbliiiing/yolov4-pytorch
  • https://github.com/maudzung/Complex-YOLOv4-Pytorch
  • https://github.com/AllanYiin/YoloV4

3. Deploy TensorRT Engine into DeepStream

3.1 Copy this directory deepstream_yolov4 into <where_deepstream_is_installed>/deepstream-5.0/sources

3.2 Compile nvdsparsebbox_Yolo.cpp in directory nvdsinfer_custom_impl_Yolo

  $ export CUDA_VER=<CUDA version of your environment, e.g. 10, 10.2, 11, etc>
  $ make

3.3 Copy the TensorRT engine into your working directory (<where_deepstream_is_installed>/deepstream-5.0/sources/deepstream_yolov4)

4. Update config_infer_primary_yoloV4.txt and deepstream_app_config_yoloV4.txt

4.1 Key options in config_infer_primary_yoloV4.txt that you may update for your own model

• [property]

  • model-engine-file
  • labelfile-path
  • batch-size
  • network-mode

• [class-attrs-all]

  • nms-iou-threshold
  • pre-cluster-threshold

Please refer to DeepStream plugin document for more information about plugin options

4.2 Key options in deepstream_app_config_yoloV4.txt that you may update for your own model

• [tiled-display]

  • width
  • height

• [streammux]

  • batch-size
  • width
  • height

• [primary-gie]

  • model-engine-file (the same as in [property])
  • labelfile-path (the same as in [property])
  • batch-size (the same as in [property])

• [tracker]

  • tracker-width
  • tracker-height

Please refer to DeepStream plugin document for more information about plugin options

5. Run the sample

  $ deepstream-app -c deepstream_app_config_yoloV4.txt