Add jpeg adapter, support for image/jpeg mimetype #796
Conversation
In [25]: run['primary']['external']['manta-cam1'].metadata
Out[25]:
{'data_key': 'manta-cam1',
'mimetype': 'multipart/related;type=image/jpeg',
'parameters': {'chunk_shape': [1, 544, 728],
'template': 'proj_6ffba08d-9416-472c-974d-fbe5a1b1fcdd_{:06d}.jpeg'},
'run_start': '0d1ab7fc-3e95-40ac-96df-0dd4820af8a7',
'uid': '986d0ee3-1528-4e75-86c3-14f4feaba9fe',
'uri': 'file://localhost/nsls2/data/tst/legacy/mock-proposals/2024-1/pass-000000/assets/'}
In [26]: run['primary']['external']['manta-cam1'][7]
Out[26]:
array([[2, 2, 2, ..., 0, 0, 0],
[2, 2, 2, ..., 0, 0, 0],
[2, 2, 2, ..., 0, 0, 0],
...,
[0, 0, 0, ..., 0, 0, 0],
[0, 0, 0, ..., 0, 0, 0],
[0, 0, 0, ..., 0, 0, 0]], dtype=uint8)
In [27]: |
| } | ||
| IMG_SEQUENCE_EMPTY_NAME_ROOT = "_unnamed" | ||
|
|
||
| IMG_SEQUENCE_MIMETYPES = { |
There was a problem hiding this comment.
just out of curiosity, does anyone use .png? or maybe other more esoteric formats?
There was a problem hiding this comment.
I am not aware of any of our beamlines writing pngs, I think a compressed format but not as compressed as jpeg is probably of limited use - if they want actual data they'd use tiff, if they want a snapshot where they don't care if data is compressed, jpeg is probably fine
| from .type_alliases import JSON, NDSlice | ||
|
|
||
|
|
||
| class JPEGAdapter: |
There was a problem hiding this comment.
I wonder if there's a chance to define base classes, something like ImageAdapter and ImageSequenceAdapter and then subclass them for jpeg and tiff? What we will need to specify when subclassing are the supported mimetypes, and file reading functions (for a single file and for the sequence) -- otherwise the code looks almost identical.
Happy to help with restructuring it!
There was a problem hiding this comment.
Yes, I think the two can easily be combined, and probably should. Maybe just passing a callable that accepts a filepath to an image and returns an np array would be enough?
There was a problem hiding this comment.
Yes, that was what I thought too, first; for tiffs and jpegs alone, just passing a callable should be enough, but this may not extend to any other formats (or if there's need for some custom logic) in the future (this is partly why I was asking about pngs). I think, I'd be in favor of keeping seperate JPEGAdapter and TiffAdapter classes (subclassed from ImageAdapter), even if they are very short.
tiled/adapters/jpeg.py
Outdated
| self._provided_metadata = metadata or {} | ||
| self.access_policy = access_policy | ||
| if structure is None: | ||
| shape = (len(self._seq), *self.read(slice=0).shape) |
There was a problem hiding this comment.
I ran into this problem yesterday, and it took me some time to figure out what was going on... During the init time, read() method may still be partially undefined, for example if it accesses self.structure(). I know it's not the case right now, but this limits our ability to extend it later (unfortunately, in a very non-obvious way).
| ------- | ||
|
|
||
| """ | ||
| return {} |
There was a problem hiding this comment.
| return {} | |
| return self._provided_metadata.copy() |
tiled/adapters/jpeg.py
Outdated
| shape = (len(self._seq), *self.read(slice=0).shape) | ||
| structure = ArrayStructure( | ||
| shape=shape, | ||
| # one chunks per underlying TIFF file |
There was a problem hiding this comment.
| # one chunks per underlying TIFF file | |
| # one chunks per underlying JPEG file |
(although, unnecessary if we move this part to FileSequenceAdapter)
Generalizing image adapters code
Checklist
Sister PR to bluesky/bluesky#1816. Tested in the lab with a real camera.
Still needs added tests, and currently requires source install of ophyd-async branch that includes AD JPEG writer.Edit: Tests added and all passing now.