Segmentation performance on ShapeNetPart dataset is exceptional. Please prove your SOTA performance!!!

[duskNNNN]

Reproduction Environment

• OS: Ubuntu 20.04.6 LTS focal x86_64
• Kernel: 3.10.0-1160.el7.x86_64
• CPU: Intel Xeon Gold 6326 (64) @ 2.901GHz
• Memory: 67078MiB / 515467MiB
• GPU: NVIDIA RTX A6000 48G
• Cuda: 11.8
• pytorch: 1.13.1+cu117

Reproduction Results

• Best accuracy is: 0.94714
• Best class avg mIOU is: 0.84012 (which is far from 89.87)
• Best inctance avg mIOU is: 0.85706 (which is far from 90.26)

Reproduction Details

Because PointRWKV did not open source the code for the component segmentation part, I tried to reproduce it using the open source component segmentation code of Point-MAE.
Segmentation code is: https://github.com/Pang-Yatian/Point-MAE/tree/main/segmentation
Log txt: pt.txt

Requisition

The paper has been accepted. Please open all source code, otherwise it will become a high-scoring article that no one cares about.

[zhegerushigui666]

good

[slc1112]

how you reproduce it. A lot of files are missing