About the update of your code.

[dof-pikes]

I noticed that you made modifications to the code repository after several peers reported large Ped errors. In particular, you added https://github.com/azhuantou/HSSDA/blob/main/pcdet/models/detectors/pv_rcnn_ssl.py#L476.

Can you provide a rationale for making these changes? Would these modifications deviate from the content of the paper?

Thank you for your response.

[azhuantou]

Hi, thanks for your interests.

I'm not sure why some peers' reproduced results have lower Ped performance. My speculation is that the most likely reason is the generation of incorrect pseudo-labels. Therefore, I implemented this hard code to prevent such situations. This does not deviate from the original paper's idea. And I will make every effort to figure out why.

If you have any new discoveries, feel free to share them directly. Thank you!

[dof-pikes]

I'm delighted by your prompt response, thank you for your dedication.

Masking Ped implies that the method lacks generalization？
Could you provide a detailed explanation of the functionality of this line of code?
https://github.com/azhuantou/HSSDA/blob/main/pcdet/models/detectors/pv_rcnn_ssl.py#82