How to import dataset with multi label

#1

I would like to import dataset ADE20K as training/test data. How to import dataset with multi label because I need to classify multi object and its position in a scenario?

#2

Hi @Rich, we currently do not support multi-label inputs and non-categorical outputs.

#3

thank @janjongboom for quick reply. There is also no metadata(ex: object coordinate) reported, right?

#4

Correct. We’ve been working on some other models, including regression, but we don’t have a date yet for which we’ll support multi-label outputs / object detection.

#5

Any update on importing datasets with bounding box files for FOMO? I am looking at uploading some MNIST information, but would only be useful if the bounding boxes could also upload to edgeimpulse. I am having no problem exporting data sets and importing them with bounding boxes from one project to another.

#6

@Rocksetta You can already? Just get the labels into this format: Uploader - Edge Impulse Documentation and name it bounding_boxes.labels and upload while also uploading normal pictures.

1 Like
#7

Hi @janjongboom. I am wondering if multi-label and non-categorical outputs are still not supported! Would you have any suggestion to address multi-task learning on a microcontroller? Maybe multiple models?