Downloads · 30 days
0
mit67_fusion_summarize
nilnguyen2k/mit67_fusion_summarize
mit67_fusion_summarize is a image classification model from nilnguyen2k. Use it when you need a label for an image. The card lists the license as mit.
Checkpoint cho dau hop nhat da nhanh (CNN/ViT/GNN) tren MIT-67, dang phan du: logits = log(pbase) + scale delta(z). Moi cau hinh co 2 bien the: nodelta (chi to hop logits goc) va delta (co them phan hop nhat dac trung).
Access
Public
Updated Aug 23, 2026
Repo size
2.2 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt1.8 MB · 77%
At a glance
- Task
- Image Classification
- License
- mit
- Access
- Public
- Created
- Aug 23, 2026
- Updated
- Aug 23, 2026
- SHA
- 8db96684
Model card
From the Hugging Face model README
MIT-67 Fusion Head Checkpoints
Checkpoint cho dau hop nhat da nhanh (CNN/ViT/GNN) tren MIT-67, dang phan du: logits = log(p_base) + scale * delta(z). Moi cau hinh co 2 bien the: no_delta (chi to hop logits goc) va delta (co them phan hop nhat dac trung).
| Cau hinh | Test acc | Macro F1 | Weighted F1 | Gap train-test |
|---|---|---|---|---|
| CNN+VIT [attention] | 89.79 +/- 0.05 | 88.62 | 89.71 | -0.43 |
| CNN+VIT [concat] | 89.78 +/- 0.05 | 88.62 | 89.70 | -0.36 |
| CNN+VIT+GNN [attention] | 89.74 +/- 0.02 | 88.58 | 89.66 | -0.33 |
| CNN+VIT+GNN [concat] | 89.73 +/- 0.03 | 88.57 | 89.64 | -0.36 |
| VIT+GNN [attention] | 89.56 +/- 0.06 | 88.35 | 89.48 | -0.32 |
| VIT+GNN [concat] | 89.52 +/- 0.03 | 88.27 | 89.45 | -0.25 |
| VIT | 89.45 +/- 0.07 | 88.29 | 89.38 | -0.30 |
| CNN | 83.21 +/- 0.09 | 80.92 | 82.88 | +0.81 |
| GNN | 60.66 +/- 0.22 | 54.85 | 59.77 | +2.36 |
- Task
- Image Classification
- License
- mit
- Created
- Aug 23, 2026
- Updated
- Aug 23, 2026