TuSimple / centerformer

Implementation for CenterFormer: Center-based Transformer for 3D Object Detection (ECCV 2022)
MIT License
292 stars 28 forks source link

Can I reproduce your great work with MMDetection3d? #12

Closed shb9793 closed 1 year ago

shb9793 commented 1 year ago

Thanks for your great work! I have ever tried CenterPoint based on MMDetection3d. I wonder which parts were changed compared with the orignal CenterPoint. Looking forward to your early reply. Many thanks!

edwardzhou130 commented 1 year ago

Hi, thanks for your interest in our work. Compared with CenterPoint, there are three significant changes:

  1. We modified the RPN to include features from more scales.
  2. We completely changed the detection head to separate the center classification and box regression process. The CenterFormer structure is applied after the selection of center candidates. And the regression head is now on 1D center features rather than 2D BEV features.
  3. We design a multi-frame fusion method that uses a different data processing pipeline.
shb9793 commented 1 year ago

Many thanks for your details!

QingXIA233 commented 1 year ago

Many thanks for your details! @shb9793 Hello, did you manage to get CenterFormer to work on MMDetection3d yet?

shb9793 commented 1 year ago

Not yet. But I noticed that MMDetection3d official will support CenterFormer in their framework as soon as possible.

QingXIA233 commented 1 year ago

Sounds good! But I didn't see this news in the README nor in the issues. Where did you get the information?