about second stage on nuscenes

[qingzhouzhen]

I notice "Two stage refinement does not bring an improvement over the single-stage CenterPoint" in article, is there any specific experiment? for example they are exactly the same or low than one-stage? @tianweiy

[tianweiy]

actually, we found that it somewhat works in our later paper. See table 3 of https://arxiv.org/pdf/2111.06881.pdf

It is about 1 map improvement and not as significant as the improvement on Waymo probably due to the sparse point cloud in nuScenes

[qingzhouzhen]

actually, we found that it somewhat works in our later paper. See table 3 of https://arxiv.org/pdf/2111.06881.pdf

It is about 1 map improvement and not as significant as the improvement on Waymo probably due to the sparse point cloud in nuScenes

Thanks for your reply, but MVP is multimodal and the pointcloud is not as sparse as original point cloud after adding virtual point in nuScenes, So could you kindly share me CenterPoint's second stage performance in nuScenes ?

[tianweiy]

In that table, we have two stage for lidar-only.

[qingzhouzhen]

ok，thx

---Original--- From: "Tianwei @.> Date: Thu, Feb 16, 2023 21:37 PM To: @.>; Cc: @.@.>; Subject: Re: [tianweiy/CenterPoint] about second stage on nuscenes (Issue#380)

In that table, I have two stage for lidar-only.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[sankin97]

Have you been able to successfully replicate the two-stage performance? I’ve discovered that utilizing the ROI head for refinement results in a significant decrease in performance on the Nuscenes dataset.

[qingzhouzhen]

Yes I do，I get 1 percent increase confided as MVP with Voxelnet backbone on nuScenes.

---Original--- From: @.> Date: Sun, May 21, 2023 22:02 PM To: @.>; Cc: @.@.>; Subject: Re: [tianweiy/CenterPoint] about second stage on nuscenes (Issue#380)

Have you been able to successfully replicate the two-stage performance? I’ve discovered that utilizing the ROI head for refinement results in a significant decrease in performance on the Nuscenes dataset.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[sankin97]

How about the performance when not using MVP.