happinesslz / SEED

[ECCV 2024] A Simple and Effective 3D DETR in Point Clouds
54 stars 0 forks source link

when do you plan to release your code #1

Closed Zoewayho closed 1 month ago

Zoewayho commented 3 months ago

Thank you for your excellent work. I am very interested in it, when do you plan to release your code?

happinesslz commented 3 months ago

@Zoewayho Thanks for your attention! We will release the code before ECCV 2024 conference.

Zoewayho commented 3 months ago

I wanna know that how do you design your prediction head when you use your Topk feature to predict K coarse box, i also tried, but the loss of the coarse 3D box would not decrease after reaching a certain threshold. and it appears in the very early stage of training. I have wondered why it works on 2D but not in 3D, but now your works seems that this approach still works on 3D. let me know if it is convenient for you.

happinesslz commented 2 months ago

@Zoewayho Hi, we adopt a coarse-to-fine query selection~(Refer to the details of DQS module). Here, maybe you need note the details of the losses in foreground query selection~(coarse queries) and quality query selection~(fine queries) .

happinesslz commented 1 month ago

@Zoewayho We have released the official code of SEED.