refactor slidedata #72
Conversation
pathml/core/slide_data.py
Outdated
| assert isinstance(slide, Slide), f"slide is of type {type(slide)} but must be a subclass of pathml.core.slide.Slide" | ||
| self.slide = slide | ||
| self._slidetype = type(slide) | ||
| self.name = None if slide is None else slide.name |
There was a problem hiding this comment.
Shouldn't every SlideData have a name?
ALso how would we get a SlideData object with no slide?
There was a problem hiding this comment.
Agreed. We need to decide if we will allow SlideData with only tiles?
There was a problem hiding this comment.
Hmm. My first thought is no - if there is a collection of tiles, it could just be a Dataset class?
| if self.slide.backend == 'openslide': | ||
| if level == None: | ||
| # TODO: is this the right default for openslide? | ||
| level = 1 | ||
| j, i = self.slide.level_dimensions[level] | ||
|
|
||
| if stride is None: | ||
| stride_i = shape[0] | ||
| stride_j = shape[1] | ||
|
|
||
| n_chunk_i = (i-shape[0])// stride_i +1 | ||
| n_chunk_j = (j-shape[1])// stride_i +1 | ||
|
|
||
| if pad: | ||
| n_chunk_i = i // stride_i +1 | ||
| n_chunk_j = j // stride_j +1 | ||
|
|
||
| for ix_i in range(n_chunk_i): | ||
| for ix_j in range(n_chunk_j): | ||
|
|
||
| region = self.slide.read_region( | ||
| location = (int(ix_j * stride_j), int(ix_i * stride_i)), | ||
| level = level, size = (shape[0], shape[1]) | ||
| ) | ||
| region_rgb = pil_to_rgb(region) | ||
| coords = (ix_i, ix_j) | ||
| if self.masks is not None: | ||
| # TODO: test this line | ||
| masks_chunk = self.masks.slice([int(ix_j*stride_j):int(ix_j*stride_j)+size,int(ix_i*stride_i):int(ix_i*stride_i)+size, ...]) | ||
| yield Tile(region_rgb, masks_chunk, coords) | ||
|
|
||
| elif self.slide.backend == 'bioformats': | ||
| # TODO: this is complicated because need to handle both chunking, allocating different 2GB java arrays, and managing java heap | ||
| pass |
There was a problem hiding this comment.
Since the exact implementation will be different depending on the Slide backend, would it be cleaner to handle that logic in the Slide class?
There was a problem hiding this comment.
Should we move the whole generate_tiles method to the Slide class? I think that would be logical. Pseudocode of what I'm thinking would happen when you run a pipeline:
def run(self, pipeline):
for tile in self.slide.generate_tiles(...):
tile_processed = pipeline(tile)
self.tiles.add(key, tile_processed)
closes #71
TODO: