Scaleup in LetterBox

[Li-Yidong]

Hi,
I have a doubt about scaleup in LetterBox. When doing training by default settings, will scaleup be True or False?

[UltralyticsAssistant]

👋 Hello @Li-Yidong, thank you for reaching out about Ultralytics 🚀! We suggest checking the Docs for answers to common questions, which cover Python and CLI usage.

For questions about the scaleup setting in LetterBox, it would be helpful to share a bit more context, such as the training configuration you are using. This will assist our engineers in providing a more accurate response. If you suspect there's a bug 🐛, please include a minimum reproducible example to help us investigate.

Join our vibrant Ultralytics community for real-time discussions on Discord 🎧. For more in-depth conversation, Discourse and our Subreddit are great places to connect and share knowledge.

Upgrade

Ensure you're using the latest ultralytics package by upgrading your installation:

pip install -U ultralytics

To avoid any compatibility issues, use Python>=3.8 and PyTorch>=1.8.

Environments

YOLOv8 can be run in several pre-verified environments:

• Notebooks with free GPU:
• Google Cloud with GCP Quickstart Guide
• AWS with AWS Quickstart Guide
• Docker with Docker Quickstart Guide

Status

Currently, all Ultralytics CI tests are passing, ensuring the correct operation of all YOLOv8 Modes and Tasks.

This response is automated, but an Ultralytics engineer will help you soon! 😊

[glenn-jocher]

@Li-Yidong by default, scaleup is set to True during training in LetterBox.

[Li-Yidong]

@glenn-jocher Thanks for your response. Is it possible that set scaleup to False during traing?

[glenn-jocher]

@Li-Yidong yes, you can set scaleup to False during training by modifying the augmentation settings in your configuration.

[Li-Yidong]

@glenn-jocher I added scaleup: False into yolov8m-obb.yaml and its runing well. But I didn't find scaleup in https://github.com/ultralytics/ultralytics/blob/main/ultralytics/cfg/default.yaml. Is it just a document issue or something else?
The yaml file is as below:

# Ultralytics YOLO 🚀, AGPL-3.0 license
# YOLOv8 Oriented Bounding Boxes (OBB) model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect
 
# Parameters
nc: 4  # number of classes
scales: # model compound scaling constants, i.e. 'model=yolov8n.yaml' will call yolov8.yaml with scale 'n'
  # [depth, width, max_channels]
  n: [0.33, 0.25, 1024]  # YOLOv8n summary: 225 layers,  3157200 parameters,  3157184 gradients,   8.9 GFLOPs
  s: [0.33, 0.50, 1024]  # YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients,  28.8 GFLOPs
  m: [0.67, 0.75, 768]   # YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients,  79.3 GFLOPs
  l: [1.00, 1.00, 512]   # YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPs
  x: [1.00, 1.25, 512]   # YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOPs
 
scaleup: False
 
# YOLOv8.0n backbone
backbone:
  # [from, repeats, module, args]
  - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
  - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
  - [-1, 3, C2f, [128, True]]
  - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
  - [-1, 6, C2f, [256, True]]
  - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
  - [-1, 6, C2f, [512, True]]
  - [-1, 1, Conv, [1024, 3, 2]]  # 7-P5/32
  - [-1, 3, C2f, [1024, True]]
  - [-1, 1, SPPF, [1024, 5]]  # 9
 
# YOLOv8.0n head
head:
  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
  - [-1, 3, C2f, [512]]  # 12
 
  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
  - [-1, 3, C2f, [256]]  # 15 (P3/8-small)
 
  - [-1, 1, Conv, [256, 3, 2]]
  - [[-1, 12], 1, Concat, [1]]  # cat head P4
  - [-1, 3, C2f, [512]]  # 18 (P4/16-medium)
 
  - [-1, 1, Conv, [512, 3, 2]]
  - [[-1, 9], 1, Concat, [1]]  # cat head P5
  - [-1, 3, C2f, [1024]]  # 21 (P5/32-large)
 
  - [[15, 18, 21], 1, OBB, [nc, 1]]  # OBB(P3, P4, P5)

[glenn-jocher]

@Li-Yidong the scaleup parameter isn't included in defaults.yaml as it's specific to certain configurations. It's not a documentation issue, but rather a customization option for your specific use case.

[Li-Yidong]

@glenn-jocher Thanks!!

[glenn-jocher]

@Li-Yidong you're welcome! If you have any more questions, feel free to ask.