Optional Tools and Libraries

DeepLearningToolkit

The DeepLearningToolkit (DLTK) is a set of environments equipped with several libraries related to deep learning. It enables running protocols in Scipion that require deep learning tools, available for both GPU and CPU. The installer will automatically detect your configuration. Since xmipp-3.25.06-Rhea all environments except micCleaner (tensorflow-gpu<2 needs anaconda) do not require any private channels (non-free channels for commercial purposes).

Note

For faster and more efficient dependency resolution in Conda environments, it is highly recommended to use LibMamba. In Conda 23 and later, LibMamba is enabled by default. For earlier versions, it can be installed and configured manually to achieve faster environment solving

Requirements

NVIDIA drivers version 450 or higher are required.

  • To check your NVIDIA driver version, run:

    nvidia-smi

  • If an older version is detected, DLTK will be installed without GPU support.

How to Install

To install DLTK, run:

scipion3 installb deepLearningToolkit

If the installation would take over 30 minutes, please accelerate the process using the libmamba solver, which can provide up to a 4x speed improvement. Note that libmamba is only available for Conda >=4.12 and must be installed and configured in your Conda environment. For more information, refer to the official guide: A Faster Conda for a Growing Community

List of Environments

  • xmipp_DLTK_v0.3 (GPU version)

Protocols using this environment: resolution_deepres, screen_deepConsensus

- python=3.7
- libstdcxx-ng
- tensorflow==1.15.0
- numpy==1.19.5
- h5py==2.10.0
- keras==2.2.5
- scikit-learn==0.22
- protobuf==3.20.3

  • xmipp_DLTK_v1.0 (GPU version)

Protocol using this environment: deep_center_predict Program using this enviroment: deep_misalignment_detection

- python=3.8
- libstdcxx-ng
- tensorflow==2.7
- numpy==1.22
- absl-py<0.12
- protobuf==3.20.3

  • xmipp_MicCleaner

Protocols using this environment: deepMicrographScreen

- python=3.9
- libstdcxx-ng # To be compatible with scipion env
- cudatoolkit=11.8
- cudnn=8.9.*
- pip
- pip:
  - micrograph-cleaner-em==1.0.2

  • xmipp_deepEMhancer

Protocols using this environment: protocol_deepEMhancer

- python=3.10
- deepemhancer=0.16 #CUDA > 10 required. Drivers nvidia >= 450.80.02
- numba
- h5py
- libstdcxx-ng

  • xmipp_pyTorch (GPU version)

Protocols using this environment: deepHand, classify_pca

- python=3.10
- numpy=1.23
- mrcfile=1.4.3
- kornia=0.6.8
- starfile=0.4.11
- requests=2
- pillow
- libstdcxx-ng
- pytorch-gpu=1.13.1
- torchvision=0.14.0
- cudatoolkit=11.3
- libcurl=7

Matlab

Xmipp has a binding to MATLAB, which allows the user to run specific Xmipp functions inside MATLAB. It is required to have a regular MATLAB installation.

Make sure you have these settings in your configuration file (xmipp-bundle/xmipp.conf) before compiling Xmipp:

XMIPP_USE_MATLAB=ON

MATLAB_DIR=<path to your MATLAB instalation> (usually something like: MATLAB_DIR=/home/user/MATLAB/R2021b)

Run

  1. Compile Xmipp normally (once the settings are as above): ./xmipp or scipion run ./xmipp

  2. Open MATLAB

  3. In MATLAB, set the path to Xmipp binding: HOME > Set Path > Add Folder... and select the path to the binding (<path to xmipp>/xmipp-bundle/build/bindings/matlab), then, click in Open and Save

  4. Now you should be able to run functions like xmipp_read() in MATLAB