Gregor Kobsik

In Table 2 you given some insights on the 'peak memory usage' per device with a batch size of 32. Do you refer to an effective batch size of 32...

I would be strongly interested in your results. As far as I understand it, we would have to have a similar function as `corresponding_points_alignment`, but more like `corresponding_points_normals_alignment`. That way...

A custom color background allows to define a seamleas background in the printing, although the corners of the images still introduce artifacts which are not desireable.  A full crop...

another possibility would be to use an shrink operator on the whole image, to contract it by 1px or 2px from the outer edge, and hopefully removing those artifacts.

It seems like the provided models need to be called with different evaluation scripts. `eval.py` uses the `Chair_dgeo` dir and `eval_recon_sn.py` uses `Chair_dhier`. I start to suspect, that the derivation...

**Example** (Shape No. 2197): ``` torch.Size([1, 10]) tensor([[-0.4375, -0.4406, -0.2583, 0.0892, 0.0740, 0.0314, 0.6235, -0.3335, -0.6235, 0.3335]]) ``` ``` torch.Size([12]) tensor([-4.3752e-01, -4.4511e-01, -2.5151e-01, 3.1384e-02, 6.2768e-02, 6.6692e-02, -1.0000e+00, 1.8370e-16, 0.0000e+00, 1.8370e-16,...

It looks like the computet/provided dataset does differ when it comes to rotation and size: After inspecting the dataloaders, it seems like the first representation uses a quaternion and the...

Although, the boxes seem to fit the point clouds: 

To fix the issue, I added the commented lines of code back. And also changed the used dataset in `eval.py` from the training set to the test set. Can you...

It seems to me, that the provided code is identical with the predecessor of the paper (StructureNet): https://github.com/daerduoCarey/structurenet/blob/master/code/vis_pc.ipynb