search

johnrobinsn / blog_comments

0 stars 0 forks source link

posts/05_ssd_final/ #4

Open utterances-bot opened 7 months ago

utterances-bot commented 7 months ago

Object Detection - Part 5 Final

Predicting multiple objects in an image

https://www.storminthecastle.com/posts/05_ssd_final/

dajcs commented 7 months ago

Great stuff, thank you for sharing :-) This was the only example of working code with fastai object detection. I could learn a lot trying to reproduce for one of our CV homeworks, the only problem is that I have a custom dataset, where I have only images and bbox coordinates. I was looking how to put these coordinates into a COCO style json, but that format has a huge amount of fields. Is there a way to feed only the bbox coords, or if I'm going with the COCO style - do I need to fill in all those fields, or there is a bare minimum where torch won't complain?