maskflow-fiji: Object Detection and Segmentation

A Fiji plugin for object detection and tracking based on Mask RCNN.

It allows object detection, segmentation and tracking using a pre-trained model with the associated maskflow Python package.

IMPORTANT: This plugin is a proof of concept and in a very early stage. At the moment only a simple model to detect microtubules is provided. I hope to be able to continue the development at some point.

Usage

• Start Fiji.
• Click on Help ▶ Update....
• In the new window, click on Manage update sites.
• Scroll to find Maskflow in the column Name. Click on it.
• Click Close and then Apply changes.
• Restart Fiji.
• Open your image.
• Run the commands at Plugins ► Maskflow.

Screenshots

Scripting

Here is an example script:

# @Dataset data
# @CommandService cs
# @ModuleService ms

from sc.fiji.maskflow import ObjectDetector

inputs = {"model": None,
          "modelName": "Microtubule",
          "dataset": data,
          "fillROIManager": True}}
module = ms.waitFor(cs.run(ObjectDetector, True, inputs))

table = module.getOutput("table")
masks = module.getOutput("masks")

The plugin also comes with an object tracker based on the centroid of the detected masks:

# @Dataset data
# @CommandService cs
# @ModuleService ms

from sc.fiji.maskflow import ObjectDetector
from sc.fiji.maskflow import ObjectTracker

inputs = {"model": None,
          "modelName": "Microtubule",
          "dataset": data,
          "fillROIManager": True}
module = ms.waitFor(cs.run(ObjectDetector, True, inputs))

table = module.getOutput("table")
masks = module.getOutput("masks")

inputs = {"masks": masks,
          "table": table,
          "linkingMaxDistance": 10,
          "gapClosingMaxDistance": 10,
          "maxFrameGap": 5,
          "fillROIManager": True}
          
module = ms.waitFor(cs.run(ObjectTracker, True, inputs))
table = module.getOutput("resultTable")

There is also a command that combine both detection and tracking:

# @Dataset data
# @CommandService cs
# @ModuleService ms

from sc.fiji.maskflow import ObjectDetectAndTrack

inputs = {"model": None,
          "modelName": "Microtubule",
          "dataset": data,
          "linkingMaxDistance": 10,
          "gapClosingMaxDistance": 10,
          "maxFrameGap": 5,
          "fillROIManager": True}
module = ms.waitFor(cs.run(ObjectDetectAndTrack, True, inputs))

table = module.getOutput("resultsTable")
masks = module.getOutput("masks")

Available Models

Objects	Version	Description	Image Size	URL
Microtubule	0.1	WIP	1280x1280	https://storage.googleapis.com/nn-models/microtubule-v0.1.zip

GPU Support

This type of neural networks are much more faster on GPU than CPU. To enable GPU support you need to manually replace libtensorflow_jni.jar to libtensorflow_jni_gpu.jar in your Fiji jars/ folder.

Authors

maskflow-fiji has been created by Hadrien Mary.

License

MIT. See LICENSE.txt