will this work on novel objects that also have novel class names that not necessarily have a lot of visual similarity with objects/classes in the training?

sailor-z / Unseen_Object_Pose

MIT License

26 stars 5 forks source link

will this work on novel objects that also have novel class names that not necessarily have a lot of visual similarity with objects/classes in the training? #4

Closed monajalal closed 1 year ago

sailor-z commented 1 year ago

I'm sorry I don't really understand. Could you please explain more about the question?

monajalal commented 1 year ago

Thanks a lot for your response.

Imagine that we have only dogs, cats, and dolphins in our training dataset. Would this be able to do inference on ducks if I provide the RGB photo of the duck as well as the 3D mesh of the duck during inference time?

sailor-z commented 1 year ago

Sure. That's our point. Our method can handle the unseen object (ducks in your case) pose estimation without retraining.