search

tudelft3d / PSSNet

PSSNet: Planarity-sensible Semantic Segmentation of Large-scale Urban Meshes
GNU General Public License v3.0
36 stars 6 forks source link

step-2 training #2

Closed ziwenjie closed 1 year ago

ziwenjie commented 1 year ago

Hi, Doctor Gao: I am very interest in your work, the step-1 has over-segmented the urban meshes with a very high accuracy. However, I use the released code step-2 to train the sum-helsinki dataset, the classification results is very bad. I tried many leanring-rate, and the results is bad too. 微信图片_20230613122130

Best regard!

WeixiaoGao commented 1 year ago

Hi, thanks for your interest. If you keep the default parameters, you should be able to get similar results to those in the paper. So I don't think the learning rate is the problem. The problem is most likely in your input data. Can you check that the input data you got from Step 1 is correct and in the correct folder? Please refer to the closed issue.