- provides a plug-and-play deep learning solution for large-scale image segmentation of light, electron and X-ray microscopy.
- is distributed as cloud formation template for AWS cloud instances, as docker container and as singularity container for local installs or supercomputer clusters.
- is backwards compatible, allowing users to continue using models that have been trained with earlier versions of CDeep3M.
- compared to v1.6.3 provides improvements in speed for larger datasets
- facilitates additional augmentation strategies (secondary: noise additions, denoising, contrast modifications; tertiary: re-sizing)
- facilitates providing multiple training volumes to train broadly tuned models.
- provides enhanced robustness using automated image enhancements.
- code implemented in Python 3.
- Generates automatically enhanced images and an overlay of the segmentation with the enhanced images for visual verification.
| Use | Description | Link | Documentation | |
|---|---|---|---|---|
| CDeep3M2-Preview: | Extremely quick tests, fully automated instantaneous runs | Link | Documentation | |
| CDeep3M2-Docker: | Local or remote, large runs, long trainings, simple installation, GPU with min 12GB vRAM required | Link | Documentation | |
| CDeep3M2-AWS: | Remote, large runs, long trainings, simple installation, pay for GPU/hour (entry level 0.50$/h) | Link | Documentation | |
| CDeep3M2-Colab: | Remote, short runs or re-training, simple installation, free GPU access | Link | Documentation | |
| CDeep3M2-Singularity: | Local or cluster, large runs, long trainings, often required for compute cluster | Link | Documentation |
-
Training: train the CNN to recognize objects in 3D image stacks (or 2D images)
-
Transfer Learning: adapt a pre-trained model to a new dataset or different object
For prediction and transfer learning you can use either a pre-trained model from the modelzoo or your own model generated during training.