Andreas Poehlmann

- [ ] create pado dataset from urlpath with images - [ ] ingest csv as metadata into pado dataset - [ ] ingest geojson (or qpproj?) as annotations into...

We provide a `Mapping[ImageId, PadoItem]` interface with `__getitem__` and `__len__` in which we correctly (and intuitively) raise `KeyError(image_id)` Now additionally we'd like to provide a `Sequence[PadoItem]` interface, again via `__getitem__`...

This would be a step closer to allowing bootstrapping of datasets via the pado cli. dependents would register themselves via entrypoints, and the pado cli would provide a bootstrap option...

Specifically for ImagePredictions but applies equally to other provider types: Since it is unlikely that predictions will all be made at the same time, it will be useful to either...

This relates to viewers implemented to show pado image predictions. It's actually not so complicated. - [ ] write pyramidal zarr with as many dimensions as you want - [...

This is essential to allow iterative model improvement without having to recreate empty datasets.

The easiest way off the top of my head would be: - [ ] convert to PNG dzi - [ ] store PNG dzi in uncompressed tar - [ ]...

This could improve performance in cased only one datatype is needed.

```python ds = PadoDataset(None, mode="w") ds.ingest_obj(ds0) ds.ingest_obj(ds1) ``` returns an error. Temporary solution ```python ds.ingest_obj(ImageProvider(ds1.images, identifier="something_else"))) ds.ingest_obj(MetadataProvider(ds1.metadata, identifier="something_else"))) ... ``` Todo: - [ ] I need to define how two...

maybe just a function ```python def is_all_local(obj: PadoDataset | ImageProvider) -> bool: ... ``` Or provide many of those "asserts" in a submodule `pado.asserts`