Addax-Data-Science/SOCAL-IRC-v3-6

This repository contains open-source models redistributed for easy integration with AddaxAI, hosted by Addax Data Science. Each model retains its original license (see license files) and attribution. We comply with all original license terms. Users must review and comply with individual model licenses before use. See below for detailed model information including original sources, licenses, and attributions.

Developer

Irvine Ranch Conservancy (IRC), Orange County, California

Links

• Learn more
• License (CC BY-NC 4.0)
• Citation

Model

IRC Species Classification Model v3.6, trained with the MEWC framework. EfficientNetV2 Small at 384x384, Keras 3.3.3, 18 classes. The weights file IRC.Species.Classification.Model.keras is the unmodified asset of GitHub release v3.6.1 (sha256 9ca887db2dac4067a532465e6cc18e71320639123c2a3d2d8bb42ce8ba1af6af). Held-out test set of 3,669 images: overall accuracy 96.9%, macro F1 0.966 (numbers from the developer's README).

Files

• inference.py: AddaxAI ModelInference implementation (TensorFlow backend).
• class_list.yaml: the developer's class map, unmodified. Its values are the training-table order written by mewc-train, not the model's output indices; the model outputs follow the class names in alphabetical order (scikit-learn LabelEncoder), which is how inference.py and mewc-predict read them. Class names are emitted lowercase.
• taxonomy.csv: class to taxonomy mapping for the AddaxAI filter tree, resolved against GBIF.
• LICENSE: CC BY-NC 4.0, as published by the developer.