Instructions for executing the official YOLOv7 on NVIDIA Deepstream.

The YOLOv7 already has a method for exporting a trained model to an ONNX format which can be run like this:

python export.py \
--weights ./yolov7-tiny.pt \
--grid \
--end2end \
--simplify \
--topk-all 100 \
--iou-thres 0.65 \
--conf-thres 0.35 \
--img-size 640 640

This command will create an ONNX model with an efficientNMS (Non Max Supression) node inserted. Care should be taken to ensure the parameters topk-all, iou-thres and conf-thres are configured correctly for how you want to execute your model. See the efficientNMS README for further details.

At time of writing, NVIDIA does not support parsing the efficientNMS output format into the correct data structure for Deepstream NvDsObjectMeta. We are in progress of donating this code to NVIDIA so it can be included in the Deepstream releases by default so this step may not be required in future.

To be able to map the model outputs from efficientNMS to the NVIDIA NvDsObjectMeta data structure you need this code has been added in /opt/nvidia/deepstream/deepstream-6.1/sources/libs/nvdsinfer_customparser/nvdsinfer_custombboxparser.cpp. A copy of this modified file exists in this repository that you should use to replace the file in that directory.

Once replaced, run make in the /opt/nvidia/deepstream/deepstream-6.1/sources/libs/nvdsinfer_customparser directory to compile the library with the new function.

You then just need to make sure your nvinfer stages point to the correct custom library and function like:

parse-bbox-func-name=NvDsInferParseCustomEfficientNMS
custom-lib-path=/opt/nvidia/deepstream/deepstream-6.1/sources/libs/nvdsinfer_customparser/libnvds_infercustomparser.so

Messages like this indicate that at least one of the detections had a class id larger than the nvinfer configured num-detected-classes. The solution is to update num-detected-classes to the correct number of classes for your model.

Assertion `(unsigned int) p_classes[i] < detectionParams.numClassesConfigured’ failed.

The original /opt/nvidia/deepstream/deepstream-6.1/sources/libs/nvdsinfer_customparser/nvdsinfer_custombboxparser.cpp contains this license which this code retains.

Changes by VisualCortex are under the MIT License and a LICENSE file is attached.

Copyright (c) 2018-2020, NVIDIA CORPORATION. All rights reserved.

Permission is hereby granted, free of charge, to any person obtaining a
copy of this software and associated documentation files (the "Software"),
to deal in the Software without restriction, including without limitation
the rights to use, copy, modify, merge, publish, distribute, sublicense,
and/or sell copies of the Software, and to permit persons to whom the
Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.