Data Extension Request

#1
by VINAYU7 - opened

I'm currently using this 10K .PLY dataset for 3D model training on Kaggle. It's excellent, but I need significantly more data (aiming for 40,000–50,000+ training steps).

I know the 125K 3D models version is available, but the large zipped format is difficult to use due to Kaggle's storage limits.

Request: Could you please consider adding more individual, unzipped .PLY 3D models to this dataset in your kaggle dataset (If I am not wrong It's Plyverse Dataset On Kaggle)?

This update would greatly help scale up the training and resolve the compatibility issue with Kaggle's environment.

Hi, the reason I didn't upload the 125k version to Kaggle is because Kaggle dataset size limits would not allow me to upload a dataset that large (I believe that limit was 100 GB, the 10k version being 87 GB not leaving much room for much more 3D models).

Solution 1: I could potentially create multiple 100 GB public datasets on Kaggle named "Plyverse 125k part 1", etc. But each "part" would be a separate public dataset listing not one unified dataset - if that would be compatible with your workflow?

I assume however you are training the network requires that the dataset is specifically pulled from Kaggle? If you explain more the limitations you are working with maybe I can better formulate a solution for you if Solution 1 is not a viable solution.

thank you for suggesting Solution 1!

Creating multiple public datasets (like "Plyverse 125k part 1," etc.) is perfectly compatible with my workflow. My workflow easily sets multiple input paths, such as
/kaggle/input/Plyverse-Part-1 and /kaggle/input/Plyverse-Part-2,

my training process will simply pull the data from those different folders. As long as those folders contain the individual .ply 3D models ready to use, this is the ideal solution from you. kaggle gives around 214GB for private storage total limit, no sure about individual dataset's limit in that total.

Thank you so much for offering to do this!...

Thank You So much, this would be really helpful in my training. Really grateful to you✨

Sign up or log in to comment