Shalin Mehta

`viscy --help` prints a useful and succinct help message. But, `viscy subcommand --help` prints a lot of lightning CLI info that is not relevant, e.g., `viscy preprocess --help` prints: ```...

strategies to improve invariance with data augmentation

1

We use intensity scaling and noise augmentations to make the virtual staining model invariant. We should leverage augmentations while keeping the training process stable and efficient. [This](https://arxiv.org/pdf/2203.03304.pdf) paper suggests a...

architectures and nomenculature

5

Now that 2D and 2.5D UNets from our 2020 paper are implemented in pytorch, we are exploring the space of architectures in two ways: a) input-output tensor dimensions. b) Using...

draft demo dataset and notebook

1

**Dataset** HEK293T cells with phase, membrane, and nuclei channels. Let's start with 50 FOVs. **checkpoint 1** Load zarr store, view label-free and fluorescence channels, configure model, browse the 2D UNet...

sampling strategy to improve class balance across cell cycle

1

At a recent meeting, we discussed strategies to achieve class balance across the cell cycle. @ziw-liu proposed a selection of FOVs based on the rough measure of the shape of...

iohub rechunk CLI

2

During the DL course, @edyoshikun needed to re-chunk the OME-Zarr store to improve the data i/o efficiency during training. Ed, please turn your code into an `iohub rechunk .zarr` CLI...

identify optimal chunk size for computational imaging, DL, and visualization workflows

9

When data is stored in locations that have significant i/o latency (storage server on premise or on cloud), a large number of small files create excessive overhead in data transfer...

With a variety of models, we see that virtually stained structures show high-frequency fluctuations over time. This may be due to the sensitivity of ConvNets to small changes in local...

projection head (and other architectural changes)

3

Address #139, #138, #120

Update the projection head (normalization and size).

2

TL;DR: current projection head doesn't do what it is supposed to do. It should have a batch norm and the size of the features and projections may be reduced further....