This is just a car patch classification example

[lucasjinreal]

I think this repo has nothing to do with 3D box prediction but just a simple and naive classification of car image patchs?

It's make no sense to do such a job and claim that can do 3D box predictions......

[salmedina]

It dos 3D box prediction in a really constrained scenario where the direction where the cars move in the video is from/towards a vanishing point. And then it extracts the patches and classifies. I believe that for the problem the author was working on, it was good enough.

[lucasjinreal]

@salmedina Classify what? the vehicle type? I don't think that can produce the 3d box

[gushart]

@jinfagang hello. Can you explain me shortly what does happening in this repo?

[VineINDalvik]

@salmedina Classify what? the vehicle type? I don't think that can produce the 3d box

"dataset.pkl" contains the 3d annotation, and 3d object detection help fine-grained classification, which exists as a intermediate process. You can refer to this paper: https://arxiv.org/pdf/1703.00686v1.pdf