Building a dataset with Edge Impulse


I’m working on a project that has object detection block, so I used Edge Impulse to build a pilot dataset with arround 700 images, I already finished annotation with bounding boxes and I have few suggestions as a feedback:

1- To have an auto-completion while entering the name of object label based on the already entered labels. This saves the developers for looking back for typos.
2- To have an optional queue (like labeling queue) named, for example, QA queue. This will make it easier to check all images and annotations. Currently browsing them one by one from " Collected data " list is not as easy as the queue.
3- To have an option to enable/disable a label in the whole model/dataset in one click (check boxe for each label). I think the current way right now is to delete the label from all images. This will make experiments easier.
4- To have additional tools while drawing bounding boxes (clear all labels, zoom in/out, …etc).

1 Like

@yahyatawil Thanks for the feedback and we will consider your suggestions and overall how to improve the labeling experience for object detection @janjongboom @dansitu

@yahyatawil 3) is now live. On Data acquisition, click the pie chart with your label, then click the “checkbox” icon above the items:

You can then disable all items for a certain class.


Thanks. This is very helpful. :smiley: