Custom Dataset Structure

i want to finetune sam3 on my custom dataset. I just want to ask what should be the structure of the dataset folder.
for example this is a proposed structure:
dataset_name/
├── train/
│ ├── image1.jpg
│ ├── image2.jpg
│ └── ...
├── valid/
│ ├── image1.jpg
│ ├── image2.jpg
│ └── ...
├── test/
│ ├── image1.jpg
│ ├── image2.jpg
│ └── ...
├── train_annotations.json
├── valid_annotations.json
└── test_annotations.json
should the annotations be in the dataset folder, or inside another folder called annotations? or inside each train, test, val?

The issue names no file, test, or entry point. Start by locating the SAM 3 fine-tuning dataset loader and any existing dataset documentation, then verify the expected image splits and annotation paths. Done means documenting one unambiguous folder layout and annotation location for custom datasets.

Written by the indexing model from the issue text.