Questions about segmentation

[accoutzzz]

Thanks for your sharing! I benefit a lot from the code. However, I also meet some problems.
Based on the released segmentation code, I conducted the experiment on the S3DIS dataset and evaluated the network on Area5. The training pipeline is following the train_semseg.py
I got the result that the training_acc is 94.3% and the test_mIoU only achieve 54.3%. The result is lower than that in the paper about 7%. I have the following question.

1. I used the SGD optimizer with LR = 0.01 and used the CosineAnnealing strategy to adjust the LR every epoch. I also used the random scale and rotation around the z-axis data argument. Did I use the correct training trick?
2. When I experiment with the ShapePartNet dataset for part-segmentation tasks, I found that testing with a multi-scale strategy brings about a 5% improvement. I wonder that did you use the multi-scale test strategy on Area5 for the semantic segmentation?
3. I also found that the experiment result of DGCNN on Table4 is from 6-fold cross-validation but not Area5.

[MenghaoGuo]

Thanks for your attention.

I do not use this repo to conduct experiment on the S3DIS dataset. If you want to reproduce the result on scene segmentation. you can adopt this repo.

There are many potential reasons that may lead to performance degradation such as data enhancement and tricks. As for the specific reason, I think a detailed ablation study should be done.

[accoutzzz]

Thanks you for your reply. I will try my best to reproduce the it.

[h1063135843]

Thanks you for your reply. I will try my best to reproduce the it.

Have you reproduced it? I am also trying to do the same thing. But I get the worse performance on s3dis (48% miou). I delete the cls_label and corresponding lable_conv according to the #8 , and set the points number to 2048 according to the #22 . What is the next should I do to get the better performance.

[294coder]

收到