Through the study of fastai course, I have gained a deep understanding of classification and segmentation. However, at the present stage, the application of object detection is very wide, but I have little understanding of its relevant contents.Is there any good information to promote the learning of object detection knowledge? Thanks~
This is a great resource:
(if you only interested in the newest stuff be sure to look at part 3/4)
I found this article to be great: http://www.telesens.co/2018/03/11/object-detection-and-classification-using-r-cnns/#ITEM-1455-0
Not quite sure about your question. But I would say to look at this, it is a good start:
Hey, there’s a previous iteration of the course where Jeremy talks about it. Check this out Object detection 1 , Object detection 2 and finally papers with code. I think if you try replicate the examples and review the papers with specific goals in mind you’ll probably be able to implement it in your own use case.
I want to do fastai v2 implementations of object detection algorithms such as yolo v3 and ssd variants. Anyone interested in doing a study group around object detection algorithms?
@zevarela here’s some starter code for object detection I did (this used the SSD technique from v1, has a few issues but if you check the discussion forum there’s someone who’s been working on it, I’d focus on it for the data block )
Wow! Thanks a lot! This seems to be exactly what I was looking for!
FWIW, let me plug my own blog post on the topic. https://machinethink.net/blog/object-detection/
Many thanks @machinethink, that’s a great post!
@zevarela did anyone ever join you to do this? I’m very interested. I’ve been looking at and trying to dive into @muellerzr 's notebook for the last several days which I’m running into issues applying to a different dataset. It seems I’m not alone based on other posts and it’d be good to get your feedback and maybe we can work together.
Honestly I would recommend IceVision instead of that notebook, and eventually I’ll be updating that notebook to utilize their library. It’s much better integrated with the api
OK. I came across IceVision. I had bookmarked that to come back to but maybe I should re-prioritize. Thanks