Hello Everyone,
I would like to know have anyone tried Object Detection (or any Computer Vision related tasks) for 360 Degree Images. When projected into 2D space, the image is heavily distorted. What is the best approach for this type of images in tasks related to computer Vision? Please share your thoughts and insights on this topic. Thanks in advance.
can you show, what kind of 360° image you have (and explain, how you acquire it) ?
(e.g. equirectangular can be projected back to cube faces)
Hi,
Consider an image like this from Google Street View
Image might have been acquired and processed by a 360 Degree camera suite like Ladybug5
what “objects” specifically do you want to detect, and where in the image will they be?
just run inference on that picture. it’ll be okay.
if you need to, you could collect some training data from such views and fine tune the DNN on that.
nobody ever needs to handle a complete sphere uniformly. for autonomous driving, it’s good enough to consider a “band” of the surroundings because nothing will fly above you and you can’t look underneath the car anyway.
