How to change yolov8-obb network input size?

[saurabh-git-dev]

Is it possible to change the yolov8-obb network size to 640x640 or 480x480 or not?
I trained a model with image size of 480x480 but it did not changed.
But last month I build a pose model in which the training images' aspect ratio were different. And after training model input size changed to probably 640x384(not exactly remember but some-where close to this value).

I believe the 32 factor is something import to select sizes but not sure how internally it is handled.

[UltralyticsAssistant]

👋 Hello @saurabh-git-dev, thank you for your question about Ultralytics YOLOv8 🚀! It seems like you're exploring the network input sizes.

If you're looking to customize the input size, the Docs have great examples that might provide clarity, especially on configuring models.

If this is a 🐛 Bug Report regarding expected behavior on resizing not being applied, we suggest providing a minimum reproducible example which would aid in accurate troubleshooting.

For custom training ❓ Questions, sharing further details like dataset examples or logs can be helpful, and ensure you're checking our Model Training Tips for optimal practices.

Join our lively community discussions on Discord 🎧, Discourse, or our Subreddit to connect further with others.

Upgrade

To ensure you're on the latest version, upgrade your ultralytics package along with all requirements in a Python>=3.8 environment with PyTorch>=1.8:

pip install -U ultralytics

Environments

You can utilize YOLO in various verified environments:

• Notebooks:
• Google Cloud, Amazon AMI, and Docker options are also available.

Status

If green, tests are passing across all major environments and tasks.

This is an automated response, and an Ultralytics engineer will assist shortly to provide further insights. 😊

[glenn-jocher]

@saurabh-git-dev to change the YOLOv8-OBB input size, specify the imgsz parameter during training or inference, ensuring it's a multiple of 32. For example, use imgsz=640 or imgsz=480. For more details, please refer to our model training tips.

[saurabh-git-dev]

@glenn-jocher
I have added the imgsz. But nothing works. I also tried setting **dynamic=True and False.**
. But when i convert the model to onnx and check the network input size it shows (1, 3, 1024, 1024).

[glenn-jocher]

@saurabh-git-dev to change the input size, ensure you're setting imgsz=640 or imgsz=480 during training and export. If the ONNX model still shows (1, 3, 1024, 1024), verify the export command includes the desired imgsz. For more details, visit the Ultralytics documentation.