liuchangji

@eunseon02 Hi, did you solve your tracking problem and if you solved it how you did?

did you solve this problem?I met the same question...

ok thank you very much

在predict.py中添加 `MetadataCatalog.get("fruits_nuts").thing_classes = ["date", "fig", "hazelnut"]`

Such kind of non-unique object has infinitely many solutions. So you need to do a post-processing after getting results from foundationPose. For example, temporal filter.

Did you solve it?I meet the same problem.

> I found it useful to put the tensor on the CPU before using `kornia.geometry.transform.warp_perspective`, but it would increase the computation time. I actually want to know the minimum amount...

As so far your CAD model has no color texture.