Recommended Augmentation and Post-Process Technique for Small Object Detection

I am currently creating a dataset of flying birds, what are the best augmentation and post-process technique for this application? The dataset has 4k images where the average pixels for the annotations are 50x50 px.

Hi @quad - thanks for posting!

For large images with small objects, you’re going to want to use SAHI.

As for general augmentations, I always recommend starting with none and layering more on as tests. Especially with small objects, you run the risk of over-augmenting your images such that the annotations are unreadable.

thank you very much!

1 Like

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.