Train on Non-Square Images

[anujdutt9]

❔Question

Hi. I am trying to train the YOLOv5s on Crowd Human dataset, but instead of using the default 640x640 square image, I would like to use 512x320 image, width and height are a multiple of 32. What's the best approach of doing this?

Additional context

I have seen some issues answered here saying to use a --rect flag, but how do I pass in my image size as the train.py accepts only a single size for [train, test]?
Also, can I pass in these images as is, without using the --rect flag as I would like to use the mosaic augmentation?

Thanks

[github-actions]

👋 Hello @anujdutt9, 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]

@anujdutt9

python train.py --img 512 --rect

[anujdutt9]

Hi @glenn-jocher Thanks for your reply. I did try this and but it looks like the image has my provided width but not the height, as there is no way to provide it here. I did check out your discussion here and it looks like you take one dimension as constant and the other dimension is taken as closest as possible to nearest multiple of 32. But what if I need to keep both the image height and width as constant? Like 512x320. Both are multiples of 32 and hence the resulting model should be able to take this as input. But instead the resulting model, even after using --rect flag takes the input as 512x512.
Should I change the code to take both height and width as input instead of just one dimension?
Thanks

[timothylimyl]

Hi @anujdutt9 , may I know whether did you manage to get it to a custom height (e.g. 320)?