search

CapPow / HerbASAP

An effort to automate image post processing for natural history collections
GNU General Public License v3.0
8 stars 3 forks source link

A way to recognize or train for different color chips #35

Open jbest opened 5 years ago

jbest commented 5 years ago

We (and many other herbaria) use a different color chip so it would be nice to have a way to recognize different color chips.

See this image for an example of one used by BRIT: https://bisque.cyverse.org/image_service/image/00-Bu6svkTKkNx5hdB8niSokV/resize:4000/format:jpeg

It is a ColorTrax 2x3 card that has been cut down the middle https://www.cameratrax.com/color_balance_2x3.php

CapPow commented 5 years ago

@jbest , Good point. My understanding is UI directed transfer learning would be a pretty large expansion for the project. For now the plan has been to train it on the majority of color chips in use, so it has out of the box functionality for most collections. That training is limited to the types of chips we could find on SERNEC through automated means and I'm pretty sure this format is not included in the training data, thanks! If you know of any other collections using differing formats I'd love a list of collection codes. Included in training set I know are chips such as: This This This and This.

I opened a related project.

dakilaledesma commented 5 years ago

I will likely try and train the models on all the color chips that I can find, however, a separate GitHub containing the neural networks models (as well as the up-to-date models) will be made public to allow users to train their own models (if need be).