how to run test.py correctly and how to visualize the result?

[yaoxingzhi]

thank you for your contribution!
I have some questions about test.py, while I running this python file,the result is so weird as the following.

Loading data...
5_hallway_1_KDTree.pkl 6.2 MB loaded in 0.2s
5_hallway_15_KDTree.pkl 2.4 MB loaded in 0.0s
5_WC_2_KDTree.pkl 1.2 MB loaded in 0.0s
5_conferenceRoom_2_KDTree.pkl 3.7 MB loaded in 0.0s
5_conferenceRoom_1_KDTree.pkl 1.9 MB loaded in 0.0s
5_WC_1_KDTree.pkl 1.3 MB loaded in 0.0s
5_hallway_14_KDTree.pkl 1.4 MB loaded in 0.0s
5_hallway_12_KDTree.pkl 1.4 MB loaded in 0.0s
5_hallway_13_KDTree.pkl 2.4 MB loaded in 0.0s
5_hallway_11_KDTree.pkl 1.5 MB loaded in 0.0s
5_hallway_5_KDTree.pkl 5.1 MB loaded in 0.0s
5_conferenceRoom_3_KDTree.pkl 2.8 MB loaded in 0.0s
5_hallway_3_KDTree.pkl 2.3 MB loaded in 0.0s
5_hallway_10_KDTree.pkl 2.2 MB loaded in 0.0s
5_hallway_4_KDTree.pkl 1.9 MB loaded in 0.0s
5_hallway_2_KDTree.pkl 8.2 MB loaded in 0.0s

Preparing reprojected indices for testing
Size of training : 16
Size of validation : 0
Loading model...
Predicting labels...
Accuracy: 0.22160644829273224
Writing results...
Assigning labels to the point cloud...
wrong field dimensions
Done. Time elapsed: 2.2s

And I can not visualize the result about S3DIS task, could you please help me ?

[LeongVan]

My results are almost the same with you when finishing running test.py. Then if you want to visualize the result, you can try tensorboard --logdir runs, and check the local host printed in the terminal hints. For example, mine is as below. The visualization result is at http://localhost:6006/.

(randlanet-pytorch) root@ubuntu:~/f/RandLA-Net-pytorch# tensorboard --logdir runs
TensorFlow installation not found - running with reduced feature set.
W1116 09:06:26.744739 140686038304576 server_ingester.py:187] Failed to communicate with data server at localhost:40801: <_InactiveRpcError of RPC that terminated with:
        status = StatusCode.DEADLINE_EXCEEDED
        details = "Deadline Exceeded"
        debug_error_string = "UNKNOWN:Deadline Exceeded {created_time:"2022-11-16T09:06:26.743382497+08:00", grpc_status:4}"
>
Serving TensorBoard on localhost; to expose to the network, use a proxy or pass --bind_all
TensorBoard 2.11.0 at http://localhost:6006/ (Press CTRL+C to quit)

[yaoxingzhi]

My results are almost the same with you when finishing running test.py. Then if you want to visualize the result, you can try tensorboard --logdir runs, and check the local host printed in the terminal hints. For example, mine is as below. The visualization result is at http://localhost:6006/.

(randlanet-pytorch) root@ubuntu:~/f/RandLA-Net-pytorch# tensorboard --logdir runs
TensorFlow installation not found - running with reduced feature set.
W1116 09:06:26.744739 140686038304576 server_ingester.py:187] Failed to communicate with data server at localhost:40801: <_InactiveRpcError of RPC that terminated with:
        status = StatusCode.DEADLINE_EXCEEDED
        details = "Deadline Exceeded"
        debug_error_string = "UNKNOWN:Deadline Exceeded {created_time:"2022-11-16T09:06:26.743382497+08:00", grpc_status:4}"
>
Serving TensorBoard on localhost; to expose to the network, use a proxy or pass --bind_all
TensorBoard 2.11.0 at http://localhost:6006/ (Press CTRL+C to quit)

Thank you for your reply, but I'm afraid my problem is not like what you said. What I want to visualize is the point cloud map after the semantic segmentation,I can visulize the result on tensorboard.

[jscksy]

maybe you can check the test.py above，the two inputs' dimensions are not correct

[xiaohe3625]

maybe you can check the test.py above，the two inputs' dimensions are not correct

Thanks for the suggestion. Have you succeeded in implementing RandLA-Net-pytorch's work?

[Dali936]

i occured this question too, whats MiniDijon8.ply meaning?

[MohamedMansour89]

Hi,
the problem is in the line where write_ply happens. It tries to write the predictions with tensor size[1,40960] along with cloud with tensor size [40960,3] and there is a mismatching in the second dimension of both. You can just expand your predictions by reshaping it :
expanded_predictions = predictions.reshape(-1,1)
also do not forget to send the clouds to cpu as well in order to avoid converting the cuda device tensor to numpy :
cloud = points.squeeze(0).cpu().numpy()[:, :3]

hope this works for you

[tempted111]

Hi,
the test.py seems to only display partial results of point clouds, and the coordinates of the point cloud are different from the original point cloud.