A ground truth dataset for 3D neuron reconstruction from electron microscopy (EM) images of the fly whole-brain, created for our project FlyWire: A human-AI collaboration to map the fly connectome. For more information, please visit https://flywire.ai/. Citation FlyWire: Online community for whole-brain connectomics Dorkenwald et al. bioRxiv 2020.08.30.274225; doi: https://doi.org/10.1101/2020.08.30.274225 Dataset description cremi_realigned.tar.gz cremi_{a,b,c}_realigned.h5: re-aligned CREMI volumes (https://cremi.org/) cremi_b_realigned_new.h5: re-aligned CREMI B volume with de-novo annotation focused.tar.gz A set of densely annotated volumes covering diverse structures in the fly brain that are underrepresented in the CREMI volumes. sparse.tar.gz A set of sparsely annotated volumes covering tricky failure modes in the initial segmentation attempt. Here intracellular structures are often oversegmented conservatively to prevent merge errors at the expense of introducing some split errors. glia.tar.gz Semi-automatically generated ground truth for glia detction. A set of subvolumes from the FlyWire segmentation was sampled from which human experts classified automatically generated segments into either neurons or glia, while existing merge errors were excluded from annotation.