tensorrt 7.2.3.4 get some wrong boxes with high score,but tensort 8.2 is correct

[we0091234]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

tensorrt 7.2.3.4 get some wrong boxes with high score,but tensort 8.2 is correct
ubuntu 16.04 tensorrt 7.2.3.4 cuda 10.2

have you tried the tensorrt 7 ？

Additional

No response

[github-actions]

👋 Hello @we0091234, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.

For business inquiries or professional support requests please visit https://ultralytics.com or email Glenn Jocher at glenn.jocher@ultralytics.com.

Requirements

Python>=3.6.0 with all requirements.txt installed including PyTorch>=1.7. To get started:

$ git clone https://github.com/ultralytics/yolov5
$ cd yolov5
$ pip install -r requirements.txt

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Google Colab and Kaggle notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training (train.py), validation (val.py), inference (detect.py) and export (export.py) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@we0091234 thanks for raising this issue! Can you implement a version check with a warning that TensorRT produces incorrect boxes?

[glenn-jocher]

@we0091234 I think this is the place:

yolov5/models/common.py

Line 330 in c1249a4

import tensorrt as trt # https://developer.nvidia.com/nvidia-tensorrt-download

[glenn-jocher]

@we0091234 good news 😃! Your original issue may now be fixed ✅ in PR #6021. This PR does not address any TRT 7 issues, but it creates a version check with a warning that TRT 8 is required by YOLOv5.

To receive this update:

• Git – git pull from within your yolov5/ directory or git clone https://github.com/ultralytics/yolov5 again
• PyTorch Hub – Force-reload model = torch.hub.load('ultralytics/yolov5', 'yolov5s', force_reload=True)
• Notebooks – View updated notebooks
• Docker – sudo docker pull ultralytics/yolov5:latest to update your image

Thank you for spotting this issue and informing us of the problem. Please let us know if this update resolves the issue for you, and feel free to inform us of any other issues you discover or feature requests that come to mind. Happy trainings with YOLOv5 🚀!

[glenn-jocher]

@we0091234 also I see different ONNX opsets are used in export.py depending on TRT version. Could ONNX Opset 12 be causing the TRT 7 issues?

yolov5/export.py

Lines 268 to 273 in 407a905

	def export_engine(model, im, file, train, half, simplify, workspace=4, verbose=False, prefix=colorstr('TensorRT:')):
	try:
	check_requirements(('tensorrt',))
	import tensorrt as trt
	opset = (12, 13)[trt.__version__[0] == '8'] # test on TensorRT 7.x and 8.x

[1chimaruGin]

I tried with opset 12 and TRT 7.2.1.3, get some wrong boxes too

.

[glenn-jocher]

@1chimaruGin what happens if you try opset 11 or 13 with TRT 7?

[1chimaruGin]

@glenn-jocher
With opset 11, also has wrong boxes too.
With opset 13, can't export TensorRT.

[glenn-jocher]

@1chimaruGin ok, then it seems like we need to set a lower limit on TRT>=8.0 with YOLOv5. I'll consult the TRT export author and add a TODO to investigate

[glenn-jocher]

@imyhxy it seems like TRT 7 export causes problems (see above).

Do you think we should implement a hard requirement for TRT 8.0 export and inference with YOLOv5? I implemented a warning (but not an error) on TRT 7 inference in #6021, but this is only a warning on inference, and does not apply to export.

[imyhxy]

@glenn-jocher Hi, jocher

I found out that there are some weird issues between TRT 7 and anchor_grid variable, when the anchor_grid is expanded to the feature size, the engine will output the wrong results (looks like the memory order of anchor_grid in TRT 7 is messed). I make a workround patch to the export_engine function and will not actually affect the model, check it out in #6185

yolov5/models/yolo.py

Line 80 in fb83929

.view((1, self.na, 1, 1, 2)).expand((1, self.na, ny, nx, 2)).float()

[we0091234]

@imyhxy great job! It's ok now with tensorrt7.2.3.4

[glenn-jocher]

@we0091234 good news 😃! Your original issue may now be fixed ✅ in PR #6185 by @imyhxy. To receive this update:

• Git – git pull from within your yolov5/ directory or git clone https://github.com/ultralytics/yolov5 again
• PyTorch Hub – Force-reload model = torch.hub.load('ultralytics/yolov5', 'yolov5s', force_reload=True)
• Notebooks – View updated notebooks
• Docker – sudo docker pull ultralytics/yolov5:latest to update your image

Thank you for spotting this issue and informing us of the problem. Please let us know if this update resolves the issue for you, and feel free to inform us of any other issues you discover or feature requests that come to mind. Happy trainings with YOLOv5 🚀!

[we0091234]

@glenn-jocher yes ,It works ！

[glenn-jocher]

@we0091234 great thanks! TODO removed.

[1chimaruGin]

@glenn-jocher

While exporting TRT, this error occured.

TensorRT: export failure: version 8.0.0 required by YOLOv5, but version 7.1.3.0 is currently installed

I think this line if trt.__version__[0] == "7" need to fix.

[glenn-jocher]

@1chimaruGin we've been updating this section a lot recently. Can you git pull to make sure your code is up to date and retry your command?

[glenn-jocher]

@1chimaruGin good news 😃! Your original issue may now be fixed ✅ in PR #6235. To receive this update:

• Git – git pull from within your yolov5/ directory or git clone https://github.com/ultralytics/yolov5 again
• PyTorch Hub – Force-reload model = torch.hub.load('ultralytics/yolov5', 'yolov5s', force_reload=True)
• Notebooks – View updated notebooks
• Docker – sudo docker pull ultralytics/yolov5:latest to update your image

Thank you for spotting this issue and informing us of the problem. Please let us know if this update resolves the issue for you, and feel free to inform us of any other issues you discover or feature requests that come to mind. Happy trainings with YOLOv5 🚀!