Skip to content

Conversation

edyoshikun
Copy link
Member

This PR adds the Tarrow package wrapped with lightning. It also contains a DataLoader to get the pair of timepoints from ome-zarr.

@edyoshikun edyoshikun marked this pull request as draft January 29, 2025 23:18
@ziw-liu ziw-liu added enhancement New feature or request representation Representation learning (SSL) labels Jan 30, 2025
@edyoshikun edyoshikun marked this pull request as ready for review February 3, 2025 16:19
@edyoshikun edyoshikun requested a review from ziw-liu February 3, 2025 16:19
@mattersoflight mattersoflight self-requested a review February 4, 2025 15:46
from tarrow.data.tarrow_dataset import TarrowDataset
from torch.utils.data import ConcatDataset, DataLoader

from viscy.utils.engine_state import set_fit_global_state
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

viscy.utils.engine_state doesn't exist. Missing commit?

@mattersoflight
Copy link
Member

@edyoshikun I was able to draw a batch and construct the model!
I now realize the model encodes temporal changes into a dense representation at each time point. The output tensor has 16 channels for each pixel in the input. In other words, the model "lifts" the image into a higher dimension instead of "compressing" it into an embedding.
image

For any meaningful comparison with dynaCLR, we will need a downstream embedding model:

image

@ziw-liu ziw-liu added this to the v0.4.0 milestone Mar 27, 2025
@edyoshikun edyoshikun changed the base branch from msd_update to main August 23, 2025 19:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request representation Representation learning (SSL)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants