Automated annotation

[carbonscott]

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.

[carbonscott]

• Precompute embeddings (Solve it in Embedding calculator and exportor #5).

Remove this plans

• New multi-way query set data loader

• Export labels assigned by the automatic label-er.

[carbonscott]

• Use embedding calculator to go through all unlabeled images so that embedding of each image is obtained.
• Score those embedding by matrix-multipy pre-computed embeddings.