Links to datasets and models - HenriquesLab/DeepBacs GitHub Wiki
Shared data and models
The table below lists the datasets and models used in DeepBacs together with the identifier on Zenodo.
| Figure | Task | Species | Zenodo link | Type |
|---|---|---|---|---|
| 2B | Segmentation | S. aureus | here | Training/test data |
| 2C | Segmentation | E. coli | here | Training/test data |
| 2D | Segmentation | B. subtilis | here | Training/test data |
| 2D | Segmentation | B. subtilis | here | Training/test data + Multilabel-U-Net model |
| S2 | Segmentation | Mixed | here | Training/test data + StarDist 2D model |
| 3A | Object detection | E. coli | here | Training/test data + YOLOv2 model |
| 3B | Object detection | E. coli | here | Training/test data + YOLOv2 model |
| 4A | Denoising | E. coli | here | Training/test data + CARE 2D model |
| 4E | Denoising | B. subtilis | here | Training/test data |
| 5A | Artificial labelling | E. coli | here | Training/test data + fnet/CARE models |
| 6A | SR prediction | E. coli | here | Training/test data + CARE model |
| 6B | SR prediction | S. aureus | here | Training/test data + CARE model |
How to use the data and models
- Download the data from the repository and upload it to your Google Drive
- Get the notebooks from the ZeroCostDL4Mic repository or from the main page of this wiki and save a copy in your drive
- Follow the instructions given in the notebook to train a model or run predictions with pretrained models
Using pretrained models in Fiji
StarDist, Noise2Void and CARE come with Fiji plugins that allow for easy application of pretrained models. As prediction requires much less computational power compared to network training, this can also be done using CPUs.
How-to-guides and much more information can be found on the developers CSBDeep homepage: