Chenxi Huang
Chenxi Huang
@yaox12, I agree with you. But I think line 14 is: `loss = (cs - ct).pow(2).sum()`. Because in paper is $$l_{coral}=\frac{1}{4d^2}||C_s-C_T||^2_F$$ and Frobenius norm is $$||A||_F=\sqrt{\sum^m\sum^n |a|^2}$$ then $$||C_s-C_T||^2_F$$ should...
I run the code directly and evaluate on validation. The kitti 40 recall results are much lower. | | Easy | Moderate | Hard |-- | -- | -- |...
> Hi, thks for the high-quality code, I have solved the problem, just using the 14500 iter in the default.py, about 60 epochs on trainval(7481) set but 120 epochs on...
For 4, are you using the model released by RTM3D? If so, which model are you using (there are two models for the same backbone); if not, can you provide...
@WenlongL 可以说一下具体更改了什么吗？代码匹配指的是？
Hi, @arik1089 . I got the same problem recently when downloading the 3dod. 
5. In addition, I would like to know the specific meanings of `last_ Box` and `proposal_ Box`, `soft-token` and `contrastive`. 6. I want to know which values should correspond to...
For question 6, is it from `Analysis Unique` and `multi`?
7. it reports `last_ position` and `last_ semantic` in validation. What are their respective meanings?
8. Under what circumstances is it appropriate to stop training, or in other words, the optimal performance is generally around the epoch number