songw-zju/meta-rangeseg

The official implementation of "Meta-RangeSeg: LiDAR Sequence Semantic Segmentation Using Multiple Feature Aggregation" (RA-L with IROS 2022)

meta-rangeseg's Issues

Question about Boundary Loss

Nice work! Thank u for releasing this repo!
I would like to ask 2 questions about BoundaryLoss~

1. Does BoundaryLoss consume a lot of time in your experiments? In my experiments, computing the Boundaryloss consumes a lot of GPU memory and triples the training time.
2. I don't use the range image method to represent the point cloud, but use the BEV two-dimensional representation, could the BoundaryLoss be used to train the network?

Thanks in advance for your reply!

Good work! Looking forward to the code.

In fact, I did try to incorporate BD Loss. However, since Lite-HDSeg is not open source, I refer to this repo for the implementation.

However, in my code, it is not as good as described in Lite-HDSeg and your ablation experiments.

Would you like to share this part of the code and your (open3d) visualization code?

Pretrained Model for Single-Scan Semantic Segmentation

Hi @songw-zju, thanks for open-sourcing your work!

Regarding the single-scan LiDAR semantic segmentation version of Meta-RangeSeg on the SemanticKITTI dataset, could you share with me the checkpoint for reproduction? Thanks a lot for your help!

Questions about residual image

Hi, thank you for your great works!! 👍 :)

Could you tell me about 'residual' param?
Can I set residual to FALSE?

Expecting your great work!

Which framework did you used to implement? Mxnet or pytorch

Question about Meta-kerenl.

Thanks for your quick respond!

Recently, I have beening working on how to improve the performence of moving object segmentation by using Range-view.
I found thaht meta-kernel is a useful block to extract feature from range image. I have try to used the source code of RangeDet, but there is some wrong with the Mxnet . So I cant debug it. I wonder if you can share the code of Meta-kernel writtern by pytorch.
It is doesn't matter that if you can't share it for any inconvinience.
Sorry for any possible missunderstanding!

And thanks again!

image

can you tell me how do you draw the image in your paper, for example

Can you provide the dependencies for training?

Pretrained model

Hi, thanks for your code and I appreciate your research.
However, pretrained model in single scene does not fit to your code. (dimension issue)
Could you check and update pretrained model?

thanks.

how to use Sub-SemanticKITTI ?

hi, thanks for your open work, I am interested in how to use the dataset consisted of every 8th frame in sequence 00-10 (except 08)? If use it, where I need to modify with the code?

Meet error when inferring

• Thanks for the wonderful work.

• According to the README, I infer the model, but I meet the bug that i can't fix. It occurs

  Meta-RangeSeg/train/tasks/semantic/dataset/kitti/parser.py

  Line 245 in 542f3e5

  exec("residual_file_" + str(i + 1) + " = " + "self.residual_files_" + str(i + 1) + "[seq][index]")

• The error message is as follows
  

• I print the seq and index, and I find seq == 08 and index == 8 when the error occurs.

if self.use_residual:
            for i in range(self.n_input_scans):
                print("seq")
                print(seq)
                print("index")
                print(index)
                exec("residual_file_" + str(i + 1) + " = " + "self.residual_files_" + str(i + 1) + "[seq][index]")

• So how can I fix this error?