Visualization about samplenet

[asafmanor]

Hi @ptxaxx
During the test pass (test1) you can create a breakpoint or modify the code to visualize the point cloud.
For example, a breakpoint in compute_pcrnet_loss at line 573 will give you an option to visualize the following:

p0 = original point cloud
p1 = rotated point cloud
p1_est = p0 rotated to fit p1

Calling plot_3d_point_cloud on these should plot them ; however it might not work out of the box. IIRC, the point clouds in the PyTorch implementation are of a different shape (Bx3xN) than of the TensorFlow implementation one (BxNx3) where B is the batch size and N is the number of points. Also, I'm not sure if plot_3d_point_cloud expects a batch dimension. If not, plotting p0[i] should work for some index i ;

I hope these instructions are clear enough!

[itailang]

Dear @ptxaxx,
You may use the following function:

SampleNet/reconstruction/src/general_utils.py

Line 141 in 0db846f

def plot_3d_point_cloud(

.

[ptxaxx]

Dear @ptxaxx, You may use the following function:

SampleNet/reconstruction/src/general_utils.py

Line 141 in 0db846f

def plot_3d_point_cloud(

.

Thank you very much for your prompt reply！
Because my foundation is weak，so I don't know how to use this function；But I'm intrigued by the idea and can't wait to see the results of my own training；I follow the registration‘s readme to implement，The result obtained is as follows：
Experiment name: log/SAMPLENET64_model_best.pth
AUC = 0.9706000000000001
Mean rotation Error = 5.049964462220669
STD rotation Error = 7.385068412955796
Mean consistency Error = 0.005256387873087079
STD consistency Error = 0.0014876173671854744
done (PID=30554)

So can you give a detailed visualization step？Thank you very much, best wishes！

[itailang]

Your results seem very good.

Les't say your point cloud is my_pc. So, for visualization simply use:
plot_3d_point_cloud(my_pc).

You can configure the function's parameters to customize the plot for your preferences.

[ptxaxx]

Hi！Sorry to bother you again！
I looked at the code you wrote, but I don't know how to modify the code to output the registration result！For example, output txt, ply, pcd, etc. I hope you can help me！
Best wishs！

[itailang]

Hi @ptxaxx,

The usage is pretty simple. Load your point cloud data according to the file format you saved it to. Then, convert the loaded point cloud to an Nx3 numpy array and use it for the plotting function.

[ptxaxx]

Hi!
Thanks for your prompt reply .
I don't know how to output the final registration result of point cloud data