About the visualization

[ztbian-bzt]

About the visualization

[LucasKre]

@ztbian-bzt I added an inference script that can be used to predict the labels for a given instance in the dataset and save the output as color-coded meshes (see Readme Inferencing). Afterwards you can inspect the meshes using MeshLab or https://3dviewer.net/)

[ztbian-bzt]

I solved the above problem and thank you for your help.

[ztbian-bzt]

@LucasKre Thanks a lot for your reply.

[ztbian-bzt]

    def get_model():
          return LitDilatedToothSegmentationNetwork() 


    model = get_model()
    if use_gpu:
        model = model.cuda()
    model = model.load_from_checkpoint(ckpt_path)

This coda may lead to the error "TypeError: The classmethod LitDilatedToothSegmentationNetwork.load_from_checkpoint cannot be called on an instance. Please call it on the class type and make sure the return value is used." I change it as follows so that it won't get this error.

    model = LitDilatedToothSegmentationNetwork.load_from_checkpoint(ckpt_path)
    if use_gpu:
        model = model.cuda()

In addition, visualization causes the teeth to elongate. How can I fix this.

[supgy]

I solved the above problem and thank you for your help.

I have encountered the same problem as you, can you teach me how to make it look regular thanks a lot

[mykakus]

I have encountered the same problem as you, can you teach me how to make it look regular thanks a lot

You need to reverse normalization steps in PreTransform class (in preprocessing.py).

 # normalize coordinate
            x[:, i] = (x[:, i] - means[i]) / stds[i]  # point 1
            x[:, i + 3] = (x[:, i + 3] - means[i]) / stds[i]  # point 2
            x[:, i + 6] = (x[:, i + 6] - means[i]) / stds[i]  # point 3
            x[:, i + 9] = (x[:, i + 9] - mins[i]) / (maxs[i] - mins[i])  # centre
            # normalize normal vector
            x[:, i + 12] = (x[:, i + 12] - nmeans[i]) / nstds[i]  # normal1
            x[:, i + 15] = (x[:, i + 15] - nmeans[i]) / nstds[i]  # normal2
            x[:, i + 18] = (x[:, i + 18] - nmeans[i]) / nstds[i]  # normal3
            x[:, i + 21] = (x[:, i + 21] - nmeans_f[i]) / nstds_f[i]  # face normal

reverse

  #  coordinate
        x[:, i] = (x[:, i] + means[i]) * stds[i]  # point 1
        x[:, i + 3] = (x[:, i + 3] + means[i]) * stds[i]  # point 2
        x[:, i + 6] = (x[:, i + 6] + means[i]) * stds[i]  # point 3
        x[:, i + 9] = (x[:, i + 9] + mins[i]) * (maxs[i] - mins[i])  # centre
  #  normal vector
        x[:, i + 12] = (x[:, i + 12] + nmeans[i]) * nstds[i]  # normal1
        x[:, i + 15] = (x[:, i + 15] + nmeans[i]) * nstds[i]  # normal2
        x[:, i + 18] = (x[:, i + 18] + nmeans[i]) * nstds[i]  # normal3
        x[:, i + 21] = (x[:, i + 21] + nmeans_f[i]) * nstds_f[i]  # face normal

Keep in mind that original maxs, mins, means etc. values have be stored before normalization and when used to reverse it.