Update README.md
Browse files
README.md
CHANGED
|
@@ -13,6 +13,18 @@ A lightweight UNet with single-block levels and sliding window attention.
|
|
| 13 |
- Token efficient visual text inputs
|
| 14 |
- Variable head in the attention modules across the layers
|
| 15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
## References
|
| 17 |
|
| 18 |
- 2411.17459
|
|
|
|
| 13 |
- Token efficient visual text inputs
|
| 14 |
- Variable head in the attention modules across the layers
|
| 15 |
|
| 16 |
+
## Introspection
|
| 17 |
+
|
| 18 |
+
Reconstruction quality, from good to worst:
|
| 19 |
+
- U-docling (this repo)
|
| 20 |
+
- U-dae
|
| 21 |
+
- U-dae-nll
|
| 22 |
+
- Eq-sae-cielab
|
| 23 |
+
- Eq-sae-cielab-8
|
| 24 |
+
- vae-f16-c4-kv
|
| 25 |
+
- vae-f16-c4
|
| 26 |
+
- vae-f16-c8
|
| 27 |
+
|
| 28 |
## References
|
| 29 |
|
| 30 |
- 2411.17459
|