Test dataset (cropped from dataset 1) for AIVE benchmarking. Voxel values are inverted from original BSE signal. Spatial units are in nm. 10nm per slice.
All raw membrane predictions and AIVE processed data from Figures 1 & 2 of manuscript, showing direct comparison of results generated via different models. The trained models are RF (Random Forest), J48 (the java compatible extension of Ross Quinlan’s C4.5 classifier), MLP (Multi-Layer Perceptron), DT (Decision Table), JRip (java compatible propositional rule-based RIPPER), and PART (Projective Adaptive Resonance Theory neural network). Slice thickness is 10nm.
Raw annotations and AIVE processed data for mitochondria from Figure 3, using classifications generated by one human on consecutive days, or two U-Nets with 3D anisotropic architecture (different random seeds). Membrane predictions by a random forest model, which were used to conduct various forms of AIVE, are also provided. Slice thickness is 10nm.
