High Positive False Issue about sfa3d

[maudzung]

Great to see your work on it. I think you can set a bigger weight for the center loss component to reduce the false positive cases.
How about its performance on Waymo? Did you evaluate?

[Zhimin-C]

Thanks for your suggestion! This is the eval result in one Waymo data sequence (200 lidar bins). I do not know why easy, medium, and hard res are the same. Besides, I also try to use resnet34 and resnet50 as a backbone but the hm_loss can't converge. Have you tried to use resnet34 or 50 and how about the result? Thanks.

Zhimin Chen

[maudzung]

I only trained the resnet18-based model.
To reduce false positive cases, you can try to increase 'peak_thresh' value. By default, its value has been set to 0.2.

[SofianeB-03]

Hello @Zhimin-C , I see that you have evaluated SFA3D on Waymo dataset. I want to evaluate on Waymo too but I did not succeed. Could you tell me how did you evaluate it please? It will be very helpful for myself.
Thank you very much.

[SofianeB-03]

@Zhimin-C, if you could share me your pre-trained weights of SFA3D on Waymo that would be really super nice (I spent a lot of time trying to do this).
Thank you a lot!