something wrong with testing

[ksama777]

Hello, author, I used the data set downloaded from the website you provided to train successfully. After that, I tested the results as follows. Is this right or is this just the test results of the two examples you provided? Segmentation: 0%| | 0/100 [00:00<?, ?it/s]Round 0ation: 0%| | 0/100 [00:00<?, ?it/s] Segmentation: 0%| | 0/100 [00:01<?, ?it/s] Segmentation: 0%| | 0/100 [00:01<?, ?it/s] Segmentation: 0%| | 0/100 [00:01<?, ?it/s] Segmentation: 0%| | 0/100 [00:01<?, ?it/s] Segmentation: 0%| | 0/100 [00:02<?, ?it/s] Segmentation: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 100/100 [01:36<00:00, 1.04it/s] Each point was visited at least 2█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 100/100 [01:36<00:00, 1.39it/s]

[ksama777]

and when I open the snapshot , it is empty , what should I do ?

[Gabry993]

Hi, in test.py you can specify the path to a testing point cloud. If you didn't change it, then yes, it will run the test on one of the "dummy" point cloud available in this dataset "data/pc_id=637/". The output that your are posting suggests that yes, it is running on this example point cloud.

Concerning the snapshot looking empty, I have the same issue sometimes. In my case, zooming in and out the view (using the mouse scrolling wheel) works for me and the point cloud appears. You can also check the output files generated during the segmentation to double check that it's working properly. Also, note that the example point cloud consists of a pretty small tile, so it's possible that if you move the view too much, you may miss it. Feel free to share the snapshot if you want, so I can double check it.

[ksama777]

thank you for you answer !!! about the pc_path , after I have finished training, which path should I use to replace it ? After completing the training, I got a checkpoints folder , does it need to be changed to the format you mentioned like these ? data/ |____pc_id=20/ | |____pc.pickle | |____metadata/ | |____metadata.pickle |____pc_id=30/ |____pc.pickle |____metadata/ |____metadata.pickle

[ksama777]

thank you for you answer !!! about the pc_path , after I have finished training, which path should I use to replace it ? After completing the training, I got a checkpoints folder , does it need to be changed to the format you mentioned like these below ? I mean can't I test that dataset directly after training ?  But there is something missing in the checkpoints folder such as 'pc.pickle' , 'metadata.pickle'   data/ |____pc_id=20/ | |____pc.pickle | |____metadata/ | |____metadata.pickle |____pc_id=30/ |____pc.pickle |____metadata/ |____metadata.pickle

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2022年5月5日(星期四) 晚上11:47 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [idsia-robotics/RandLA-Net-pytorch] something wrong with testing (Issue #6)

Hi, in test.py you can specify the path to a testing point cloud. If you didn't change it, then yes, it will run the test on one of the "dummy" point cloud available in this dataset "data/pc_id=637/". The output that your are posting suggests that yes, it is running on this example point cloud.

Concerning the snapshot looking empty, I have the same issue sometimes. In my case, zooming in and out the view (using the mouse scrolling wheel) works for me and the point cloud appears. You can also check the output files generated during the segmentation to double check that it's working properly. Also, note that the example point cloud consists of a pretty small tile, so it's possible that if you move the view too much, you may miss it. Feel free to share the snapshot if you want, so I can double check it.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>