Qingwen Zhang

> > > > Great detective work! I have been trying to figure out why I keep running out of GPU memory, and this seems to be the problem. I...

Hi, If you want to evaluate your own dataset, you need to have flow ground truth available which may hard to get. But if you only want to use pretrained...

Hi, the dataset link download is not available. Please give full access: You need access. hard to tell if I cannot open the data you provided, the conclusion we summarise...

I think you should look more detail on clean map:  Based on point level evaluation, the conclusion is still the same, since there are lots of dynamic points octomap...

If you like, I can upload the model I trained in the whole dataset as the paper set. More discussion posts I publish here and screenshot here: [https://github.com/Kin-Zhang/carla-expert/discussions/2](https://github.com/Kin-Zhang/carla-expert/discussions/2) As WPH-commit...

Here is the [onedrive download link](https://hkustconnect-my.sharepoint.com/:u:/g/personal/qzhangcb_connect_ust_hk/ETe4OX79JldIsvgNGIs5NmIBkJnfPDG2G5Ij-8ha518rOw?e=BG1dek) for model I trained in the whole dataset as the discussion said above.

Not sure why pip install --no-build-isolation have following issue: ``` (seflow) ➜ DeFlow git:(feature/benchmarks) ✗ pip install FastGeodis --no-build-isolation Collecting FastGeodis Using cached FastGeodis-1.0.5-cp38-cp38-linux_x86_64.whl Requirement already satisfied: torch>=1.5.0 in /home/kin/mambaforge/envs/seflow/lib/python3.8/site-packages...

Their dataset paper indicates the LiDAR specification:  And here is my visualization (different color means different LiDARs I also show other four LiDARs) from `scene id: 10203656353524179475_7625_000_7645_000, timestamp: 1522688015070129`...

@fwilliams also wait.... Think this two issues should be linked together: https://github.com/nv-tlabs/NKSR/issues/28

Maybe it's because of `points_per_second` default one too small? not sure.. I haven't tested yet. Guess through the CARLA official LiDAR