MtReNIRS pretrained encoder
This repository holds the pretrained MtReNIRS encoder by Chae et al. (2026), as released by its authors. The file
mtrenirs_10s.pt is a byte-identical copy of models/weights/mtrenirs_10s.pt in
microsoft/MtReNIRS at commit
16c3082
(git blob 33a6f2ee8893548ccc83db7044d93d2d6ca4550f). It is mirrored here for braindecode's MtReNIRS model
(braindecode/braindecode#1564) and its tests.
| file | sha256 |
|---|---|
mtrenirs_10s.pt |
0960d02f9437d86f0f1a670eace8905d58b33f3394e58d5c17a6451bbcb9484d |
Licence
MIT, Copyright (c) Microsoft Corporation. LICENSE.txt is the source repository's licence file, unchanged. The
authors' model card states that the pretrained encoder weights are released in that repository under the MIT
License.
Input the weights expect
10 Hz HbO/HbR in 100-sample (10 s) windows, each channel z-scored over its run and clipped at 6, with channel positions in Standard 10-05 space. The authors' model card describes the intended and out-of-scope uses.
Use with braindecode
The checkpoint holds {"config", "model"}; its keys are the braindecode model's own, and load_state_dict takes
the file as released.
import torch
from huggingface_hub import hf_hub_download
from braindecode.models import MtReNIRS
path = hf_hub_download("braindecode/mtrenirs-pretrained", "mtrenirs_10s.pt")
model = MtReNIRS(chs_info=raw.info["chs"], n_outputs=2, n_times=100, sfreq=10.0)
model.load_state_dict(torch.load(path, map_location="cpu", weights_only=True))
Citation
@article{microsoft2026mtrenirs,
author = {Chae, Tony and Avramidis, Kleanthis and Emmanouilidou, Dimitra and Tashev, Ivan J.},
title = {{MtReNIRS}: A Self-Supervised Foundation Model for Transferable Neurovascular Representations},
journal = {arXiv preprint arXiv:2610.36159},
year = {2026},
}