DeepImageJ is a user-friendly plugin that enables the use of a variety of pre-trained neural networks in ImageJ and Fiji. The plugin bridges the gap between developers of deep-learning models and end-users in life-science applications. It favors the sharing of trained models across research groups and could have a broad impact in a variety of imaging domains. DeepImageJ does not require any deep learning expertise or any computer programmer skills.

DeepImageJ has been updated to DeepImageJ 3.0. The format of the pre-trained models are compatible with the format of the Bioimage Model Zoo. Contact us if you have any question!


DeepImageJ trained models

U-Net Pancreatic Segmentation

Binary SegmentationPhase Contrast

Data: Cell Tracking Challenge
Traning: deepImageJ & I. Arganda-Carreras

U-Net Glioblastoma Segmentation

Binary SegmentationPhase Contrast

Data: Cell Tracking Challenge
Traning: JoĂŁo Luis Soares Lopes (EPFL)

MU-Lux (CTC) PhC-C2DL-PSC cells

Instance SegmentationPhase Contrast

F. Lux & P. Matula, arXiv, 2020

FRU-Net sEV Segmentation

Instance SegmentationTEM

Estibaliz GĂłmez-de-Mariscal et al., Science Reports, 2019.

DEFCoN density map estimation

Density estimationSMLM

Baptiste Ottino et al. (EPFL)

Widefield TxRED Super-resolution

Super-resolutionFluorescence

Hongda Wang et al., Nature Methods, 2019.

U-Net HeLa Segmentation

Instance SegmentationDIC

Data: Cell Tracking Challenge
Traning: JoĂŁo Luis Soares Lopes (EPFL)

Widefield DAPI Super-resolution

Super-resolutionFluorescence

Hongda Wang et al., Nature Methods, 2019.

Widefield FITC Super-resolution

Super-resolutionFluorescence

Hongda Wang et al., Nature Methods, 2019.

Jones Virtual Staining

Virtual LabellingLight Transmission

Yair Rivenson et al., Nature biomedical engineering, 2019.

MT3 Virtual Staining

Virtual LabellingLight Transmission

Yair Rivenson et al., Nature biomedical engineering, 2019.

A first selection of state-of-the-art models from various groups has been made available in the BioImage Model Zoo. Beyond its direct use, we expect deepImageJ to contribute to the spread and validation of deep learning models in life-science applications.


New functionalities of deepImageJ 3.0

  • Thanks to the Java Deep Learning Library (JDLL), on-the-fly installation of the DL engines in ImageJ/Fiji.
  • This new version of deepImageJ allows the creation of image analysis pipelines with multiple Deep Learning steps, using different frameworks.
  • The connection between all models from the Bioimage Model Zoo and Fiji is ensured from now on.
  • The prediction on deepImageJ 3.0 is usually faster than deepImageJ 2.
  • Thanks to ImgLib2, deepImageJ 3.0 handles larger images.
  • DeepImageJ 3.0 compatibility here:

    • Prediction PyTorch 1 PyTorch 2 TensorFlow 1 TensorFlow 2 ONNX
    Linux Yes *GPU Yes *GPU Yes *GPU Yes *GPU Yes
    Windows Yes *GPU Yes *GPU Yes *GPU Yes *GPU Yes
    Mac Intel Chip x86_64 Yes Yes Yes Yes *Access Yes
    Mac M1/M2 arm64 Yes Not yet Never Not yet *Java8 Yes
    Notes
    *GPU: Use GPU if everything is well configured on the machine.
    *Access: Allow to access to Files and Folders (Privacy Settings).
    *Java8 It can't work while Fiji is running on JVM Java 8! One need to wait for an update of Fiji.

References

  • Cite the original paper of the model which is bundled into deepImageJ.
  • E. GĂłmez-de-Mariscal, C. GarcĂ­a-LĂłpez-de-Haro, W. Ouyang, L. Donati, E. Lundberg, M. Unser, A. Muñoz-Barrutia, D. Sage, "DeepImageJ: A user-friendly environment to run deep learning models in ImageJ", Nat Methods 18, 1192–1195 (2021). DOI: https://doi.org/10.1038/s41592-021-01262-9
  • Access and read the paper online.
  • W. Ouyang, F. Beuttenmueller, E. GĂłmez-de-Mariscal, C. Pape, T. Burke, C. Garcia-LĂłpez-de-Haro, C. Russell, L. Moya-Sans, C. de-la-Torre-GutiĂ©rrez, D. Schmidt, D. Kutra, M. Novikov, M. Weigert, U. Schmidt, P. Bankhead, G. Jacquemet, D. Sage, R. Henriques, A. Muñoz-Barrutia, E. Lundberg, F. Jug, A. Kreshuk, BioImage Model Zoo: A Community-Driven Resource for Accessible Deep Leaning in BioImage Analysis, BioRxiv (2022). DOI: https://doi.org/10.1101/2022.06.07.495102

FAQ

The Image.sc Forum is the main discussion channel for deepImagej, hence we recommend to use it for any question or curisity related to it. Use a tag such as "deepimagej" so we can go through your questions. You will find already an extended Q&A post about Machine Learning and deepImageJ: [NEUBIAS Academy@Home] Webinar “Machine Learning/Deep Learning/DeepImageJ” + QUESTIONS & ANSWERS .


Integrations with deepImageJ


Getting close to the community

External tutorials & courses that use deepImageJ




Ecole Polytechnique Fédérale de Lausanne
Initiative EPFL Center for Imaging
Biomedical Imaging Group


GitHub repository


Universidad Carlos III de Madrid
Bioengineering and Aerospace Engineering Department