NVlabs / wetectron

Weakly-supervised object detection.
Other
362 stars 45 forks source link

Clarity on this repository #84

Closed ChrisRawstone closed 1 year ago

ChrisRawstone commented 1 year ago

Hey, I am doing a bachelor thesis on this project on some data without bounding box labels. So the data has images and labels only.

Can this github repository do the following:

image + label -> bounding box coordinates

In other words: Can this repo's model take in images and only labels without bounding boxes and spit out bounding boxes?

Can this repository do this? Sorry if this is clear but we just want to make sure.

Best regards Christian and August

ChrisRawstone commented 1 year ago

Hey all. The answer to the above is yes. You just have to generate the bounding boxes yourself using SS. We got this repository working on our custom dataset. Just in case anyone is wondering!