search

SamsungLabs / tr3d

[ICIP2023] TR3D: Towards Real-Time Indoor 3D Object Detection
Other
141 stars 8 forks source link

What is label2level in TR3Dhead? #21

Closed fangchuan closed 7 months ago

fangchuan commented 7 months ago

Hi, thanks a lot for this outstanding work! currently, I am working on training my dataset using TR3D, when I adapt this code base to my dataset, I find there is a network configuration parameter that varies from dataset to dataset. More specifically, it is the label2level in xxxx_config.py of each dataset, I notice that different scene data use different label2level, So, what is it? Why are these configs different? And does it really matter when I want to get more accurate detection results for my dataset?

https://github.com/SamsungLabs/tr3d/blob/fa82f458a01c7ee0f24f21fd021b89a81223caeb/configs/tr3d/tr3d_sunrgbd-3d-10class.py#L21

https://github.com/SamsungLabs/tr3d/blob/fa82f458a01c7ee0f24f21fd021b89a81223caeb/configs/tr3d/tr3d_scannet-3d-18class.py#L21

@filaPro thanks in advance !

filaPro commented 7 months ago

Hi @fangchuan , Please follow #9.

fangchuan commented 7 months ago

Hi @fangchuan , Please follow #9.

That's helpful, many thanks!