We have provided training logs for 6 areas on S3DIS as well as the results of 6-fold cross-validation. Due to the limitation of the submission file size, we only provide the parameter file for DSConv-L on area5. Since we used the PointNeXt code for training, the displayed parameter files and training logs are named as PointNeXt, but our algorithm was actually used. There is an error in the test area displayed in the .csv files, please refer to the folder names for accuracy. If you want to directly view all training results, it is recommended to refer to the test results of the 6-fold cross-validation. Our RTX 4070 graphics memory is insufficient to support the testing of some areas, so the 6-fold cross-validation was measured on RTX 4090 24GB. Therefore, some areas do not have .csv result files. To view the test results for those areas, please refer directly to the 6-fold cross-validation results.
