bruAristimunha commited on
Commit
92eed00
·
verified ·
1 Parent(s): 318df75

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +65 -7
README.md CHANGED
@@ -1,14 +1,72 @@
1
  ---
2
  library_name: braindecode
3
- license: apache-2.0
 
4
  tags:
5
- - DIVER1
6
  - braindecode
7
- - model_hub_mixin
 
 
 
 
8
  - pytorch_model_hub_mixin
9
  ---
10
 
11
- This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
12
- - Code: https://braindecode.org
13
- - Paper: [More Information Needed]
14
- - Docs: https://braindecode.org/stable/generated/braindecode.models.DIVER1.html
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  library_name: braindecode
3
+ license: mit
4
+ pipeline_tag: feature-extraction
5
  tags:
 
6
  - braindecode
7
+ - DIVER-1
8
+ - ieeg
9
+ - seeg
10
+ - eeg
11
+ - foundation-model
12
  - pytorch_model_hub_mixin
13
  ---
14
 
15
+ # DIVER-1, 0.1 s Tiny (iEEG)
16
+
17
+ The released encoder of [Han et al. (2025)](https://arxiv.org/abs/2512.19097),
18
+ the iEEG encoder DIVER-1-0.1s Tiny (patches of 0.1 s at 500 Hz, pretrained on intracranial recordings), 12.7M parameters, converted to
19
+ [`braindecode.models.DIVER1`](https://braindecode.org/stable/generated/braindecode.models.DIVER1.html).
20
+
21
+ Source: `weights/ieeg_pretrained_weights.pt` of [DIVER-Project/DIVER-1](https://github.com/DIVER-Project/DIVER-1/tree/25638eb38ef297b582ab79ae1c96260f57c155b3),
22
+ sha256 `b812093779fb5cf1b76e18f8307df4136b67b008b6cd94c6be75fa286a87864c` (the GitHub LFS object is no longer served; the same file is in
23
+ the authors' [Google Drive folder](https://drive.google.com/drive/folders/1Wmv36jifjE0Jj6noGOFFsqRzK-4xnj1b)). Conversion casts the bfloat16
24
+ tensors to float32, renames them and drops the pretraining-only mask token and
25
+ reconstruction heads; `convert_diver1_weights.py` in this repository reproduces
26
+ it. Encoder features match the official model to 0.0e+00 in
27
+ float32 on CPU, for intracranial and scalp channels, with muP attention scaling
28
+ (`mup_attention=True`).
29
+
30
+ ## Usage
31
+
32
+ Resample to 500 Hz and pass the electrode positions in `chs_info` (metres, as in
33
+ MNE). The channel kinds set the modality: SEEG, ECoG and DBS are intracranial,
34
+ EEG is scalp.
35
+
36
+ ```python
37
+ import torch
38
+ from braindecode.models import DIVER1
39
+
40
+ model = DIVER1.from_pretrained(
41
+ "braindecode/DIVER-1-0.1s-tiny", chs_info=raw.info["chs"], n_times=500, n_outputs=2
42
+ )
43
+ logits = model(torch.randn(8, len(raw.ch_names), 500))
44
+ ```
45
+
46
+ `n_times` must be a multiple of 50. The saved geometry is 6
47
+ channels and 500 samples; the encoder does not depend on it, so
48
+ pass your own montage, `n_times` and `n_outputs` when loading.
49
+
50
+ ## Limitations
51
+
52
+ These files hold the encoder only; the release has no classification head.
53
+ Braindecode initializes the head on load, so it needs fine-tuning, as in the
54
+ paper's protocol (`scripts/finetune_neuroprobe.sh`). The check above covers
55
+ float32 CPU encoder features, not downstream accuracy, GPU kernels or mixed
56
+ precision.
57
+
58
+ ## Citation
59
+
60
+ ```bibtex
61
+ @article{han2025diver,
62
+ title={DIVER-1: Scaling intracranial EEG foundation models for transferable representations},
63
+ author={Han, Danny Dongyeop and Gwon, Yonghyeon and Lee, Ahhyun Lucy and Lee, Taeyang and Lee, Seong Jin and Choi, Jubin and Lee, Sebin and Bang, Jihyun and Lee, Seungju and Park, David Keetae and Yoo, Shinjae and Chung, Chun Kee and Cha, Jiook},
64
+ journal={arXiv preprint arXiv:2512.19097},
65
+ year={2025}
66
+ }
67
+ ```
68
+
69
+ ## License
70
+
71
+ The weights are released by the DIVER Project under the MIT licence (`LICENSE`).
72
+ Braindecode's code is BSD-3-Clause.