added cristae model and single channel transfrom #100
Conversation
synapse_net/inference/util.py
Outdated
| if with_channels: | ||
| # TODO Check that this is the correct axis. | ||
| input_volume = torch_em.transform.raw.standardize(input_volume, axis=(1, 2, 3)) | ||
| input_volume = np.stack([torch_em.transform.raw.normalize(input_volume[0]), input_volume[1]], axis=0) |
There was a problem hiding this comment.
Here you assume that we never want to normalize the first channel, and you assume that we always have only two channels. I don't think we should make such a general assumption.
synapse_net/inference/util.py
Outdated
| model: Optional[torch.nn.Module] = None, | ||
| verbose: bool = True, | ||
| with_channels: bool = False, | ||
| channels_to_normalize: Optional[List[int]] = [0], |
There was a problem hiding this comment.
Optional implies that this value can be None. What happens if it is None?
I think in this case all channels should be normalized (independently).
I would also set this as the default, and for the cristae model pass channels_to_normalize=[0].
synapse_net/inference/util.py
Outdated
| # TODO Check that this is the correct axis. | ||
| input_volume = torch_em.transform.raw.standardize(input_volume, axis=(1, 2, 3)) | ||
| for ch in channels_to_normalize: | ||
| input_volume[ch] = torch_em.transform.raw.normalize(input_volume[ch]) |
There was a problem hiding this comment.
Let's leave this at normalize.
synapse_net/inference/util.py
Outdated
| # If we have channels then the standardization is done independently per channel. | ||
| if with_channels: | ||
| input_volume = input_volume.astype(np.float32, copy=False) | ||
| channels_to_standardize = None |
There was a problem hiding this comment.
Now you are over-riding your input so that it's always None. If you remove this line then everything should be fine.
|
Looks good and I will go ahead and merge it. (Though it doesn't yet add the model to the UI ;), but we can do this in a separate PR) |
@constantinpape