Tiling (preprocessing) or sahi (inference) or do both?

🤝 Community Help
1

Hi everyone,

I’m working on a small object detection task and trying to get the best performance possible. I’ve seen some conflicting recommendations about whether to:

  • Use tiling (preprocessing) only during training
  • Use SAHI only during inference
  • Or do both

Has anyone experimented with this? What’s the most best approach for small object detection?

Thanks!

2

Hi @Marc!
Fantastic question! The best approach for small object detection is to use SAHI during inference.

Here is a great blog post with further explanation: How to Detect Small Objects: A Guide