search

IGLICT / MeshPooling

Code for 'Mesh Variational Autoencoders with Edge Contraction Pooling'
MIT License
32 stars 7 forks source link

data processing #6

Open nainsheng opened 1 year ago

nainsheng commented 1 year ago

During the reproduction process, regarding the dataset, only in the scape dataset can MATLAB code be executed. Other datasets such as swing, face, and horse cannot be processed by MATLAB code, and size issues will be reported. May I ask if you need to preprocess these data?

FrozenSilent commented 1 year ago

Can you provide specific error information, and are these datasets from official websites?

nainsheng commented 1 year ago

Our datasets are all downloaded from the official website, but errors still be reported when we run MATLAB code. The image we provide shows error information. 图片 图片

FrozenSilent commented 1 year ago

The first issue is that the mesh itself has problems, it may not be a manifold. The second issue is that there are redundant values in the face attribute in the obj file. The obj file needs to be in the format of f v1 v2 v3 and should not exist \\. It is recommended to save the first file, i.e. 1.obj, using meshlab, excluding UV and normal information when saving.

nainsheng commented 1 year ago

Hi, thank you for your reply. We preprocessed the dataset according to the method you provided, but this type of problem may occur during the execution of Python. Looking forward to your reply. 图片

FrozenSilent commented 1 year ago

This error is caused by loading the previous dataset partition file .dat, but the new dataset does not have the same amount of data as the previous dataset. It is recommended to delete (or rename) the original dataset partition file after making changes to the dataset. The new dataset is randomly divided in the first training, and subsequent training will load the previous configuration.

zynn056 commented 1 year ago

Regarding the face and camel datasets in the paper, I traced them back to the paper you cited and successfully downloaded the relevant datasets. However, the number of vertices in the face and camel datasets I downloaded does not match the number of vertices provided in your paper. So I would like to ask if you have done any other operations on these two datasets？ Can you provide a connection between the face and fat camel dataset?

FrozenSilent commented 1 year ago

We do some preprocessing to the camel dataset. Here is the link to the camel dataset we used in our experiments. It should be noted that the code should also work on the original dataset. For the face dataset, we cannot provide it directly due to some reasons.

Noxx23 commented 5 months ago

Hi, i'm stuck on the preprocessing of my custom dataset . I assumed the .obj should have a ratio of (vertices 1: faces 2) and be watertight but i didn' t figure it out how to do it. Do you have a script ?