MEC EM Myelin Segmentation Dataset

This dataset contains 3D electron microscopy (EM) volumes and their corresponding myelin segmentation masks, from the MEC (medial entorhinal cortex) sample used in the EMNeXt project.

Content

The dataset contains 12 paired samples (indices 1 to 12). Each sample has two files:

File	Description	Typical shape	Typical dtype
{i}_raw.npy	Raw EM image volume	(125, 1250, 1250)	float32
{i}_seg_Myelin.npy	Myelin segmentation mask (instance/label IDs) for the same volume	(125, 1250, 1250)	uint16 (some samples use uint8)

Notes:

• Shapes are approximately (125, 1250, 1250); a few samples (e.g. 5, 6) are (125, 1251, 1251) instead. Always check array.shape after loading rather than assuming a fixed size.
• The dtype of *_seg_Myelin.npy is uint16 for most samples, but uint8 for a couple of samples (e.g. 8, 12) — the label range still fits in that dtype.
• raw volumes are float32 grayscale EM intensities; seg_Myelin volumes are integer label maps where each nonzero value marks myelin instance/class regions (background is typically 0).

Loading example

import numpy as np
from huggingface_hub import hf_hub_download

raw_path = hf_hub_download(repo_id="haoyuan7/mec", repo_type="dataset", filename="1_raw.npy")
seg_path = hf_hub_download(repo_id="haoyuan7/mec", repo_type="dataset", filename="1_seg_Myelin.npy")

raw = np.load(raw_path)
seg = np.load(seg_path)
print(raw.shape, raw.dtype, seg.shape, seg.dtype)

License

Released under cc-by-4.0. Please cite/credit the source project (EMNeXt / MEC dataset) when reusing this data.