About the model train result (S3DIS)

[songjiehan]

Hello,I use the model to train S3DIS, but the result is bad. The miou is 35.2.

There is only difference between my S3DIS dataset and the one in paper authors, that is, I did not follow the paper authors to obtain the point clouds with the lowest probability in order according to color probality. I am passing in one index at a time and then getting the room point cloud for that index and then culling 40960 points afterwards, may I ask if this will also lead to lower results for me?

I would like to know if you have some test results on the dataset for me to compare.

I would like to get some answers from you,my emain is 979871258@qq.com

[sun421-cloud]

Hello,I use the model to train S3DIS, but the result is bad. The miou is 35.2.

There is only difference between my S3DIS dataset and the one in paper authors, that is, I did not follow the paper authors to obtain the point clouds with the lowest probability in order according to color probality. I am passing in one index at a time and then getting the room point cloud for that index and then culling 40960 points afterwards, may I ask if this will also lead to lower results for me?

I would like to know if you have some test results on the dataset for me to compare.

I would like to get some answers from you,my emain is 979871258@qq.com

Hi, I also encountered this situation, have you solved it?