carbonscott / specklenn Goto Github PK

Leveraging models trained on a small fraction of experimental datasets for data annotation.

Use hit-labeler (https://github.com/carbonscott/hit-labeler) for checking the automated annotations.

• Provide examples about training on real experimental data.

Implement an embedding calculator. It's a task out of #2 .

Input:

• A pre-trained model.
• A label file.

Computation:

• Compute image embedding for all labeled images.
• Categorize image embedding based on their associated labels.
• Obtain the mean embedding per category.

Output:

• Export the mean embedding per category.

Substantial change in /reg/data/ana03/scratch/cwang31/spi/labels/json2csv.py, potentially affecting how deepprojection parses label file.

Transfer the feature from image or panel to mosaic mode.

Change seqi to event_num in the following cases:

https://github.com/carbonscott/deepprojection/blob/6de8d95d91da7b699cd2ecbb4ca716d4aff04567/deepprojection/datasets/images.py#L486

The message exported from the embedding calculator was written one entry at a time, which slows down embedding calculator.

• Change how it will be integrated in the dataset class.