ONNX and Pytorch model don't make same mask weight in yolov5-seg

[hsysfan]

Search before asking

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

Question

Hi, i have some questions on yolov5-seg model

first, i coded the algorithm for segmentation inference on C# OnnxRuntime

i made the image to tensor and insert the tensor to loaded model in Pytorch(Python) and OnnxRuntime(C#) for proof

i know output[1] has [1,32,160,160] as 32 masks information. So i coded the algorithm is process_mask in yolov5 to C#

In the way, i found model throw different result. please watch the images

the left is onnxruntime mask after process_mask algorithm and the right one is pytorch result.

there are some differences pixel values. maybe 0% ~ 20% difference

i checked the masks_in [1,32] <- i think this is weight per mask

<style> </style>

PyTorch	0.14222	-0.64795	0.34971	0.37515	0.74815	-0.25121	-0.9901	-0.71582	-0.09188	-0.7556	0.17439	0.06452	-0.94477	0.09694	0.7804	0.21439	0.18145	-0.60947	-0.04956	0.15793	-0.17822	-0.78358	0.54038	-0.07782	-0.51345	0.35098	-0.32386	-0.12711	0.06213	0.15394	-0.60719	-1.26161
C#	0.1513574	-0.65539	0.355499	0.378068	0.756042	-0.23867	-0.96947	-0.71049	-0.07348	-0.7692	0.196966	0.087532	-0.94167	0.080687	0.766626	0.186181	0.161458	-0.59977	-0.05438	0.15459	-0.18784	-0.79042	0.561843	-0.05658	-0.52831	0.362876	-0.31567	-0.11423	0.080159	0.16219	-0.58261	-1.263
Difference Rate(%)	6.03696945	1.135081	1.628305	0.771766	1.04387	5.253587	2.127851	0.749605	25.03916	1.768211	11.46192	26.29017	0.329692	20.14317	1.796651	15.15169	12.38252	1.616948	8.869921	2.160884	5.123703	0.865902	3.820038	37.53892	2.812743	3.278175	2.593572	11.27559	22.49123	5.086393	4.219429	0.110055424

these are weights in mask information [1,32], upper is pytorch the below is onnxruntime.

i checked all my codes, but i cant find difference on algorithm.

could you explain why the results have difference weights?

thanks for reading

Additional

No response

[github-actions]

👋 Hello @hsysfan, 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 a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Requirements

Python>=3.7.0 with all requirements.txt installed including PyTorch>=1.7. To get started:

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

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):

• 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, validation, inference, export and benchmarks on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

Introducing YOLOv8 🚀

We're excited to announce the launch of our latest state-of-the-art (SOTA) object detection model for 2023 - YOLOv8 🚀!

Designed to be fast, accurate, and easy to use, YOLOv8 is an ideal choice for a wide range of object detection, image segmentation and image classification tasks. With YOLOv8, you'll be able to quickly and accurately detect objects in real-time, streamline your workflows, and achieve new levels of accuracy in your projects.

Check out our YOLOv8 Docs for details and get started with:

pip install ultralytics

[hsysfan]

Sorry i solved my problem.

the problem was converting color image to grayscale.

my task is always using grayscale image so i converted all the image color to grayscale in my test program.

this is so basic problem.

im so sorry for confusing you

[glenn-jocher]

@hsysfan glad to hear you solved the issue! Converting the color image to grayscale makes total sense now. No need to apologize, we're all here to learn, and I'm glad I could be of help. If you have any more questions or run into any issues in the future, don't hesitate to reach out. Good luck with your project!

[sun0518]

@hsysfan Hi,when i use onnxruntime (C#) to run onnx model(seg) which is trained by yolo,the result only output center x,y and Hight&width,no box and mask corrdinate,can you help to share your C# code for this problem,thanks