Conversion from torch model to Onnx model causing variations in inferencing input parameters about ultralytics HOT 2 OPEN

Hello! It looks like you've encountered issues regarding input parameters and output resizing when converting from a PyTorch .pt model to an .onnx format. Here are some insights to help address these:

1. Input Parameters (task and imgsz): When exporting to ONNX and using it, you might need to explicitly specify these parameters that are automatically handled in the PyTorch version. This difference often comes due to the static nature of ONNX which needs definite input sizes and tasks specified.

2. Output Resizing and Padding: ONNX models sometimes behave differently due to the static graph constraints. ONNX requires fixed dimensions (imgsz) for all operations within the model, which might lead to additional resizing and zero-padding not present in the dynamical PyTorch version.

To remedy resizing and padding issues, you can adjust the post-processing of the output in your application to align with how the .pt model handles these aspects. Here is sample code to adjust the output:

# Assuming `results` is the output of the ONNX model
import numpy as np

def resize_masks(masks, original_shape):
    out = [cv2.resize(mask, (original_shape[1], original_shape[0]), interpolation=cv2.INTER_LINEAR) for mask in masks]
    return np.array(out)

original_shape = cropped_image.shape[:2]  # Assuming cropped_image shape is HxWxC
resized_masks = resize_masks(results.masks, original_shape)

Please refer to the Ultralytics documentation on model export for details on how configurations may affect ONNX outputs compared to PyTorch. If the issue persists, consider using our support channels where we can examine your case more closely. Let us know if this was helpful! 🚀

from ultralytics.

👋 Hello @liuy129, thank you for your interest in Ultralytics YOLOv8 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

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.

Join the vibrant Ultralytics Discord 🎧 community for real-time conversations and collaborations. This platform offers a perfect space to inquire, showcase your work, and connect with fellow Ultralytics users.

Install

Pip install the ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8.

pip install ultralytics

Environments

YOLOv8 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 Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLOv8 Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

from ultralytics.