CDeep3M

CDeep3M provides a plug-and-play cloud based deep learning solution for image segmentation of light, electron and X-ray microscopy.

Quickstart CDeep3M on the cloud

Click launch button to spin up the latest release of CDeep3M on the cloud (~20 minute spin up time): (Oregon region)

NOTE: Running will result in EC2 charges (0.9-3$ per hour runtime)

First time users

Sign up for AWS Account

Just opened your AWS account? Request access to GPU nodes before starting: follow instructions here

SSH key

Follow the instructions on how to link your SSH key. You can directly create the SSH key on AWS.

Launch cloudformation stack

Once approved, launch cloudformation stack using the launch button. Click here for detailed instructions on launching CDeep3M. NOTE: Running CloudFormation stack requires AWS account and will result in EC2 charges (0.9-3$ per hour runtime)

Access your cloud

Click here for instruction how to access your cloudstack

Once you launched the stack:

• Click here for instructions of a CDeep3M demorun 1

  Running segmentation with a pretrained model (Runtime ~5min)

• Click here for instructions of a CDeep3M demorun 2

  Running short training and segmentation using data already loaded on the cloud (Runtime ~1h)

• How to train your own model and segment with CDeep3M

  This will guide you step-by-step through training a network and the prediction on your own data.

Shutting AWS cloud down

Done with your segmentation? Don't forget to delete your Cloud Stack

Additional info for more experienced users

• How to retrain a pretrained network
• How to speed up processing time
• How to insert and use a validation dataset

Hyperparameters can be adjusted by passing flags to runtraining.sh

References

If you use CDeep3M for your research please cite:

@article{,
  title={CDeep3M - Plug-and-Play cloud based deep learning for image segmentation},
  author={Haberl M., Churas C., Tindall L., Boassa D., Phan S., Bushong E.A., Madany M., Akay R., Deerinck T.J., Peltier S., and Ellisman M.H.},
  journal={Nature Methods},
  year={2018}
  DOI = {10.1038/s41592-018-0106-z}
}

Further reading:

• CDeep3M open access article in NatureMethods
• CDeep3M preprint
• CDeep3M was developped based off a convolutional neural network implemented in DeepEM3D

Support

Please email to cdeep3m@gmail.com for additional questions.

Local install using Docker

Thanks to CrispyCrafter and Jurgen for making a Docker version of CDeep3M. If you want to run CDeep3M locally this should be the quickest way:

• Docker version of CDeep3M

Local install, for advanced users/developers only

Installation requirements for local install

NOTE: Getting the following software and configuration setup is not trivial. To try out CDeep3M it is suggested one try CDeep3M in the cloud, desribed above, which eliminates all the following steps.

• Nvidia K40 GPU or better (needs 12gb+ ram) with CUDA 7.5 or higher

• Special forked version of caffe found here: https://github.com/coleslaw481/caffe_nd_sense_segmentation

• Linux OS, preferably Ubuntu with Nvidia drivers installed and working correctly

• Octave 4.0+ with image package (ie under ubuntu: sudo apt install octave octave-image octave-pkg-dev)

• hdf5oct: https://github.com/stegro/hdf5oct/archive/b047e6e611e874b02740e7465f5d139e74f9765f.zip

• bats (for testing): https://github.com/bats-core/bats-core/archive/v0.4.0.tar.gz

• GNU Parallel

• Python 2.7 with cv2 (OpenCV), joblib and requests

How to install locally

Step 1) Download release tarball

wget https://github.com/CRBS/cdeep3m/archive/v1.6.3rc3.tar.gz

Step 2) Uncompress

tar -zxf v1.6.3rc3.tar.gz
cd cdeep3m-1.6.3rc3

Step 3) Add to path

export PATH=$PATH:`pwd`

Step 4) Verify

runtraining.sh --version

License

LICENSE for CDeep3M

For contents of model/ see model/LICENSE file for license

Acknowledgements

• CDeep3M was developped based off a convolutional neural network implemented in DeepEM3D

• Support from NIH grants 5P41GM103412-29 (NCMIR), 5p41GM103426-24 (NBCR), 5R01GM082949-10 (CIL)

• The DIVE lab for making DeepEM3D publicly available.

• O. Tange (2011): GNU Parallel - The Command-Line Power Tool, ;login: The USENIX Magazine, February 2011:42-47.

• This research benefitted from the use of credits from the National Institutes of Health (NIH) Cloud Credits Model Pilot, a component of the NIH Big Data to Knowledge (BD2K) program.