We have an exciting new expedition in our Labs project. It’s called LightningBug Text Correct and we’ll ask you to utilize the relatively new Text From Subject task. The basic idea is to look at the label image as well the OCR output and correct any issues you see. 

We hope it will be a fun and relatively straightforward task. These labels are unique in that they are not typical label images, but label reconstructions. Unlike many insect specimen images these images are taken with the insect and labels still on the pin using multiple cameras and angles. These label reconstructions are basically models of the labels based on several camera angles taken using the LightningBig system. The text is then read using Google’s OCR tools. You can learn more about the project on their website https://www.lightningbug.tech/ .

Our job now is to assess how well the text came out and see where we can make improvement to the label text itself. For example in the label below there would be no corrections since the text was captured correctly. Note that the object coming out between the 7 and 0 is the specimen pin. These images are captured while the specimen and labels are still on the pin so you will this frequently in this dataset.

You can give it a try on our Notes from Nature Labs project.