The pytorch tutorial for object detection finetuning includes instance segmentation, which requires segmentation mask annotations. Is there a tutorial for the (presumably simpler) task of object detection without segmentation, where the task is to learn bounding boxes with class labels only?
Thanks in advance!
I don’t believe that there is a tutorial for detection, per se, but pytorch
does provide two pure detection models in torchvision:
(I haven’t used either of these.)
Pytorch’s object-detection tutorial illustrates the use of Mask R-CNN,
but does also talk about Faster R-CNN some. So you ought to be
able to use that tutorial for Faster R-CNN (or RetinaNet), leaving
out the instance-segmentation stuff.
It does indeed talk about Faster R-CNN, but doesn’t show how to use it. I actually find easier to build a script from scratch based on the doc files than to factor out the segmentation stuff, which is disappointing given the tutorial title.
I think it’s very good for segmentation stuff, and I’m very grateful for the efforts that have been invested in this and other tutorials. Still, a short tutorial on plain object detection would be most welcome.
Hi! I feel the same way. Did you found some simple tutorial just for get me started?