search

songw-zju / Meta-RangeSeg

The official implementation of "Meta-RangeSeg: LiDAR Sequence Semantic Segmentation Using Multiple Feature Aggregation" (RA-L with IROS 2022)
MIT License
45 stars 6 forks source link

Pretrained Model for Single-Scan Semantic Segmentation #6

Closed ldkong1205 closed 1 year ago

ldkong1205 commented 1 year ago

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!

songw-zju commented 1 year ago

Hi @ldkong1205, thanks for your interest in our work.

Since I have modified some parameters in network for easy understanding, the old single-scan checkpoint cannot be used. I'm so sorry that due to being busy with other projects recently, I don't have enough time and GPUs for retraining. After that I will update the checkpoint for the single-scan LiDAR semantic segmentation as soon as possible.

ldkong1205 commented 1 year ago

Thanks for the update!

songw-zju commented 1 year ago

Hi @ldkong1205, sorry for the late reply. I have retrained the single scan network with the clean code in this repository and update the pre-trained weights with the predictions on valid/test set.

Considering that the best-performing model on the valid set will have certain fluctuations when it is submitted for the test set evaluation, our results here are slightly lower than those in the original paper.