Hi, thank you for releasing the code base!

I encountered a strange issue. When I generate point clouds, my results are different depending on if I do it sequentially, or as a batch. Here is a small example (running inside model_test.py):

with torch.no_grad():
    batch_size = 10

    # generate random latent matrices for 10 shapes
    noise = self.noise_generator(bs=batch_size)

    # First try passing noise through model one at a time (sequentially)
    x = self.sphere_generator(bs=1)
    pcs_sequential= np.zeros((batch_size, 2048, 3))
    for i, latent_matrix in enumerate(noise):
        out_pc = self.G(x, latent_matrix[None])
        out_pc = out_pc.transpose(2, 1)
        out_pc = out_pc.cpu().detach().numpy()
        out_pc = normalize_point_cloud(out_pc)
        pcs_sequential[i] = out_pc 
    
    for i in range(batch_size): # Render results
        pyplot_draw_point_cloud(pcs_sequential[i], output_filename=f'nobatch_{i}.png')

    # Now try running all noise through model (full batch) 
    x = self.sphere_generator(bs=batch_size)
    
    out_pc = self.G(x, noise)
    out_pc = out_pc.transpose(2,1)

    pcs_batching = out_pc.cpu().detach().numpy()
    pcs_batching = normalize_point_cloud(pcs_batching)
    print (pcs_sequential - pcs_batching) # <--- We should expect this to be all zeros
    
    for i in range(batch_size): # Render results
        pyplot_draw_point_cloud(pcs_batching[i], output_filename=f'batch_{i}.png')

Here the point clouds generated sequentially and the point clouds generated in a batch are different. When I compute pcs_sequential - pcs_batching the values are non-zero, and the rendered point clouds are visibly different.

Any help would be appreciated!

hello
I want to train sp-gan on my own point cloud data,what kind of dataset shoud I prepare?what is the data form of your h5 file?

Hi, thanks for sharing your implementation.
It looks like the per point discriminator loss described in the main paper is disabled.
In my understanding, the output will be (N+1)x1.
I am wondering that this is intended or?

https://github.com/liruihui/SP-GAN/blob/main/Generation/Discriminator.py#L110

Hi, Thanks for your excellent work.
Could you please share the SMPL and SMAL datasets you used?
I would be very appreciate if you can help me.

Hi, Thanks for your excellent work and the source code.
I am looking for the method for visualize the point cloud as a mesh, but I meet some problem when applying the method found
on the internet due to the lack of normal. (e.g. Poisson, ball pivot)
And I find the decent result in the bottom of Fig.9 in your paper.

Is it possible to release the source code of this part (convert point cloud to mesh)?
It will be really helpful to me.

Thank you very much!

Hi,thank you for your great work!
I have tried part regions editing by your code ,but it seems that the mask[512,1024...] can't find the correspond part shape,such as chair’s back, legs, etc.Did I have to find the points representing the chair's legs on the ball manually?How do you solve it ?

It seems that there are some issues with the code of model.py. Could you please help me to solve them?
Thanks!

thanks for release the training and visualization code, based on your advise I use the PointFlow/metrics to evaluate, but the CD and EMD both 1，the all reasults as follow picture ,I can‘t analyze why, could you give me further advise or could you release the test code? Tanks a lot.

Hi, thanks for releasing the code.

During training, SP-GAN uses a different data normalization approach (instance-wise normalizing the points to fit a unit ball) comparing to prior works such as PointFlow (zero-mean per axis and unit-variance globally). How about during the evaluation, how is the data preprocessed?

Would you release the evaluation code, or share how you preprocess/post-process the point cloud before evaluation?

Thank you!

Hi, thank you for the great code! I have some issues with the train.py file. This error appears on personal training data set once the h5 files have been created and read:

ImportError: cannot import name 'chamferdistcuda' from 'metrics.CD_EMD.cd' (unknown location)

in file SP-GAN-main\metrics\CD_EMD\cd\chamferdist\ChamferDistance.py

Thank you for your time!