search

xchhuang / simple-pytorch-3dgan

A simple and unofficial 3D-GAN implementation using PyTorch [NeurIPS 2016]
85 stars 31 forks source link

how to use other datasets? #7

Open mamantaroking opened 10 months ago

mamantaroking commented 10 months ago

You wrote that the dataset was taken from modelnet right?

So, I downloaded modelnet40 from kaggle, since modelnet is saved in an object file format (.off), i tried using a simple file converter I got from chatgpt to convert the .off file format into matlab (.mat) file format.

Let alone the part where it won't train, i printed out the content of one of my converted .mat file and one of the provided .mat dataset, and saw that the provided dataset seems to have a lot of numerical 0 which confuses me.

First, please forgive me since I am a big newbie. Second, how or where can i get the dataset that this model could train on? If you would be so kind, could you tell me the steps on how to recreate the .mat files that you have provided, and how can it be done if one were to use the modelnet40 .off files?

mamantaroking commented 10 months ago

Okay, so I found that you have included a downloadable folder of 3D shape nets in the Data section of the readme. Now the modelnet40 problem has been resolved.

But now i wonder how i can use my own dataset to train it on. If you have any idea, I would be much appreciated.

Harshilpatel17 commented 4 months ago

please provide me another Dataset Link contains MAT file for train the model

xchhuang commented 1 month ago

Sorry for the late response. As mentioned by @mamantaroking , you can download the ModelNet dataset from kaggle.

I have also added a script src/convert.py to convert .off to .mat for the kaggle ModelNet dataset. It seems working, but I haven't tried full training on this dataset.