The results is lower than your paper's and your pretrained weights

[EchoAmor]

Hello , I trained the car model by myself and didn't modify your code,but the results is 5 lower than the results in your paper and your pretrained weights, do u know where the problem is? Is there any metric I don't use? Hope for your reply,thanks!

BTW, what does the instance DB mean? I am a little confused, your paper didn't mention it

[ilbash]

@EchoAmor instance DB mean Database for Instance Segmentation for Spatial Embedding Training (Kitti INStanse dataset). What about weights - I have a different situation, mentioned there https://github.com/detectRecog/PointTrack/issues/13