Different results on object detection when inputting original img vs pre-resized img(matched to model's trained img size) about ultralytics HOT 2 OPEN

👋 Hello @tk-ryu, 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.

Hi there! Thanks for reaching out with your question.

The difference you're seeing in the results when using the original versus pre-resized images is likely due to how the aspect ratio and feature details are preserved during the resizing process. Even though the model automatically resizes inputs to match its training size, the manner in which you pre-process and resize images could be slightly different from the internal resizing process used during training or prediction by the model. This can affect how the model interprets and detects objects in the image.

For instance, subtle differences in aspect ratio or interpolation methods used during resizing can alter the pixel values and, consequently, the model's predictions. It's especially impactful when your original image size and model input size have significantly different aspect ratios or dimensions.

To minimize these discrepancies, ensure that the resizing method you use for pre-processing images exactly matches the one used during the model's training. You might also want to experiment with different interpolation methods if you are handling them manually.

Here's a code snippet to demonstrate resizing using PyTorch's functional interfaces, which may closely align with internal operations:

import torch
import torchvision.transforms.functional as F

def resize_image(img, target_size):
    # Resize the image
    img = F.resize(img, target_size)
    # Add padding if needed (to match your specific use-case)
    return img

# Example usage:
# img_tensor should be your image tensor
# resized_img = resize_image(img_tensor, (256, 416))

This ensures that the image resizing is as consistent as possible with the model's expectations. Hope this helps clarify the difference you are experiencing!

from ultralytics.