Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,32 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
**Model Card for LEAP distilled ViT-S and ViT-Tiny**
|
| 6 |
+
|
| 7 |
+
This is the ViT-Tiny and ViT-S checkpoints(No Register) distilled from ViT-G DINOv2 on ImageNet-100 and ImageNet-1K, respectively.
|
| 8 |
+
The knowledge distillation process follows the procedure proposed in "LEAP: Layer-skipping Efficiency via Adaptive
|
| 9 |
+
Progression for Vision Transformer Distillation"
|
| 10 |
+
|
| 11 |
+
Repository: https://github.com/KevinZ0217/LEAP
|
| 12 |
+
Paper: LEAP: Layer-skipping Efficiency via Adaptive Progression for Vision Transformer Distillation
|
| 13 |
+
|
| 14 |
+
Use cases: the ViT model will output feature maps that can be used to a variety of downstream tasks(classification, image retrieval, semantic segmentation)
|
| 15 |
+
|
| 16 |
+
Performance:
|
| 17 |
+
|
| 18 |
+
ImageNet-100:
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
ImageNet-1K:
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
|