search

WingkeungM / RFFS-Net

Code for "Beyond single receptive field: A receptive field fusion-and-stratification network for airborne laser scanning point cloud classification" in ISPRS Journal of Photogrammetry and Remote Sensing (ISPRS Journal Ph & RS).
30 stars 3 forks source link

Different results with reported ones #9

Open valenrach opened 1 year ago

valenrach commented 1 year ago

Hello. Thank you for your valuable study. I followed the same steps but got different results. You can see the results for reported and my outcomes, respectively.

Powerline -> 75.50 % - 67.09 % Low Veg. -> 80% - 79.96 % Imp. Surf. -> 90.50 % - 90.67 % Car -> 78.50% - 76.08 % Fence -> 45.50 % - 30.91 % Roof -> 92.70 % -90.93 % Facade -> 57.90 % -53.23 % Shrub -> 48.30 % -45.09 % Tree -> 75. 70 % -78.47 %

There are huge differences for Fence and Powerline classes. Could you please help me to solve the issue ?

WingkeungM commented 1 year ago

This may be because of the seed setting. Different seed settings have different network convergence conditions. You can try to choose to set different seeds to achieve faster convergence speed. You can also choose to increase the number of rounds of training to achieve similar performance.