Do you really need to use deep learning?

Deciding on the best method to perform a computer vision analysis is challenging. If you are new to image analysis, it may be hard to gauge how difficult one method would be vs. another.

This is a short guide, based on my own experiences/opinions/frustratinos, on how to decide whether to use “traditional” CV methods like thresholding, .

“Traditional” CV methods include things like static thresholding, adaptive thresholding, colorspace conversions, etc.

Deep learning methods

Are you imaging things in situ or in a highly controlled setting?

If you are capturing images in the field or in the greenhouse. If you are imaging detached .

Will the images be highly consistent?

You may have a .

For example, here are several images drawn from the USDA pecan breeding program:

The layout, lighting conditions, etc. are all remarkably consistent.

Are you analyzing existing data or trying to build a tool for future use?

If you are only trying to analyze.

If you want to build a tool that can be used for future datasets (in which )

###