Downloads · 30 days
0
mindeye2-subj02
kevin-bretz/mindeye2-subj02
mindeye2-subj02 is a machine learning model from kevin-bretz. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for pytorch. The card lists the license as mit.
Trained MindEye2 checkpoints for Subject 2 of the Natural Scenes Dataset (NSD). They are the base reconstruction models for the inference-time extensions in kevin-bretz/MindEyeV2: a diffusion timepoint sweep, a VLM ca…
Access
Public
Updated Jul 9, 2026
Repo size
5.9 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pth5.9 GB · 100%
At a glance
- Library
- pytorch
- License
- mit
- Access
- Public
- Created
- Jul 9, 2026
- Updated
- Jul 9, 2026
- SHA
- b3a648ee
Model card
From the Hugging Face model README
MindEye2 — Subject 2 checkpoints
Trained MindEye2 checkpoints for Subject 2 of the Natural Scenes Dataset (NSD). They are the base reconstruction models for the inference-time extensions in kevin-bretz/MindEyeV2: a diffusion timepoint sweep, a VLM caption ablation, and Brain-Optimized Inference (BOI).
These are inference-only checkpoints — each file contains model weights (model_state_dict) only, with optimizer state removed. They contain no NSD data.
Files
| Path | Model | Description |
|---|---|---|
finetuned_subj02_1sess_1024hid_low/last.pth | Subject-2 fine-tune | Fine-tuned on 1 session of Subject 2 (hidden_dim=1024, n_blocks=4, blurry_recon=True). This is the model the extensions run on. |
multisubject_excludingsubj02/last.pth | Multi-subject pretrain | Pretrained on all NSD subjects except Subject 2; the starting point for the fine-tune above. |
Usage
# Fine-tuned Subject-2 model (what the extensions use)
huggingface-cli download kevin-bretz/mindeye2-subj02 \
finetuned_subj02_1sess_1024hid_low/last.pth --local-dir train_logs
# (optional) the multi-subject pretrain it was fine-tuned from
huggingface-cli download kevin-bretz/mindeye2-subj02 \
multisubject_excludingsubj02/last.pth --local-dir train_logs
Then run inference (see the repository README):
python recon_inference.py --model_name=finetuned_subj02_1sess_1024hid_low --subj=2 \
--n_blocks=4 --hidden_dim=1024 --blurry_recon --new_test
Data access & license
The weights are released here; the underlying fMRI/stimulus data is not and must be obtained directly from NSD by agreeing to the NSD Terms & Conditions. Base MindEye2 components (unCLIP decoder, converters) are on pscotti/mindeyev2. Please cite MindEye2, MindEye1, and the Natural Scenes Dataset.
- Library
- pytorch
- License
- mit
- Created
- Jul 9, 2026
- Updated
- Jul 9, 2026