Issues with using dataset which is not is square dimensions. about ultralytics HOT 4 CLOSED

👋 Hello @agNihit928, 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.

@agNihit928 hi there,

Thank you for your detailed report and for checking existing issues before posting. It’s great to see your thorough investigation into the matter.

The warning you encountered is indeed expected behavior. Currently, YOLOv8's training process assumes square dimensions for images (imgsz as an integer) due to the way various augmentations and transformations are implemented. This is why you see the warning and why the evaluation metrics remain consistent regardless of the rectangular dimensions you attempted to set.

To address your specific points:

1. Training with Rectangular Images: As you correctly identified, the current implementation does not support non-square dimensions directly in the training pipeline. The imgsz parameter is expected to be an integer, representing square dimensions. This is a design choice to ensure compatibility and stability across various augmentation and preprocessing steps.

2. Documentation Clarification: The documentation might indeed seem misleading regarding the flexibility of imgsz. While predict and export modes support rectangular dimensions, train and val modes do not. We appreciate your feedback and will work on clarifying this in our documentation to prevent future confusion.

3. Possible Workaround: If you need to work with rectangular images, you might consider resizing your images to the nearest square dimension (e.g., padding or cropping) before feeding them into the training pipeline. This ensures compatibility with the current implementation while allowing you to work with your desired aspect ratio.

Here’s a small snippet to help you resize your images to square dimensions using padding:

import cv2
import numpy as np

def resize_to_square(image, size=640):
    h, w = image.shape[:2]
    scale = size / max(h, w)
    new_h, new_w = int(h * scale), int(w * scale)
    resized_image = cv2.resize(image, (new_w, new_h))

    delta_w = size - new_w
    delta_h = size - new_h
    top, bottom = delta_h // 2, delta_h - (delta_h // 2)
    left, right = delta_w // 2, delta_w - (delta_w // 2)

    color = [0, 0, 0]
    new_image = cv2.copyMakeBorder(resized_image, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)
    return new_image

# Example usage
image = cv2.imread('path/to/your/image.jpg')
square_image = resize_to_square(image, size=640)
cv2.imwrite('path/to/save/square_image.jpg', square_image)

This approach ensures that your images maintain their aspect ratio while fitting into the required square dimensions.

If you have any further questions or need additional assistance, feel free to ask. We’re here to help!

from ultralytics.

Thanks a lot for the confirmation @glenn-jocher .
Also, thank you the reference code, this is gonna make my work easier in the preprocessing step.

from ultralytics.

@agNihit928 you're welcome! 😊 I'm glad the reference code will help streamline your preprocessing step.

If you encounter any further issues or have additional questions, feel free to reach out. We're here to assist!

Happy coding! 🚀

from ultralytics.