Welcome to the CeNCOOS Documentation of the UCSC IFCB Classifier

Note

This diagram shows how IFCB model data is pre-processed for classification and how classification results are interpreted.

The UCSC IFCB Classifier is a trained classification model used for classifying images collected by an Imaging FlowCytobot (IFCB). The model is trained on images collected from IFCBs deployed at the Santa Cruz Wharf (SCW) and the Power Buoy (PB) in Monterey Bay.

Note

Try the classifier out on of these Jupyter Notebooks on Google Colab:
- Classifier an IFCB Image
- Classifier a Complete IFCB Sample

The model is a depthwise Convolutional Neural Network (CNN) based on the Xception architecture [Chollet, F., 2017] with 134 layers using weights pretrained on ImageNet.

Version 1.0 - 51 distinct taxanomic classes are defined by the model. A description of the classes can be found in the classes section.

The trained model weights are available at: https://huggingface.co/patcdaniel/phytoClassUCSC/tree/main

Caution

This classifier is still under development and should be applied with a healthy dose of sketpicism. There may be regional and deployment specific biases that are not well described when applying the model to new data.

Contents:

• Usage
  • Accessing the Model
  • Running the Classifier
  • Making Sense of Results
  • Caveats
• Model Classes
  • Class Description
  • Class-Specific Thresholds
• Model Metrics
  • Evaluatation
  • F1 Score
  • Class Specific Threshold
  • Softmax Scaling Experiments