-
|
Hello all, I could not find any info about preparing my own dataset. Does this repo have that kind of tutorial? Any help will be appreciated! Thanks, ZD |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
simplest way is to use the default folder layout, make one 'dataset/train' folder, and another 'dataset/val' (and possibly test if you want a sep test set. Then make a folder for each class, name them consistently across each train / val folder and put examples of each class in its corresponding folder. There are lots of other ways to do, like build a TFDS, build a Hugging Face dataset, use Webdataset and build tar files with appropriate annotations, etc. |
Beta Was this translation helpful? Give feedback.
-
|
Hello there, Thanks for your reply. Do you guys have any docs or tutorials (different ways to prepare datasets) for that? I am new to this. Is this kind of common knowledge that assumes users have? (I have already looked into the code and know how the code handles the data. ) Thanks, ZD |
Beta Was this translation helpful? Give feedback.
simplest way is to use the default folder layout, make one 'dataset/train' folder, and another 'dataset/val' (and possibly test if you want a sep test set. Then make a folder for each class, name them consistently across each train / val folder and put examples of each class in its corresponding folder.
There are lots of other ways to do, like build a TFDS, build a Hugging Face dataset, use Webdataset and build tar files with appropriate annotations, etc.