Update README.md
Browse files
README.md
CHANGED
|
@@ -77,9 +77,19 @@ etc.
|
|
| 77 |
|
| 78 |
High quality 5000 images from danbooru. They were shuffled and split into train:eval at 4500:500. (Same as p1atdev/siglip-tagger-test-2)
|
| 79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
|
| 81 |
## Training procedure
|
| 82 |
|
|
|
|
|
|
|
|
|
|
| 83 |
### Training hyperparameters
|
| 84 |
|
| 85 |
The following hyperparameters were used during training:
|
|
|
|
| 77 |
|
| 78 |
High quality 5000 images from danbooru. They were shuffled and split into train:eval at 4500:500. (Same as p1atdev/siglip-tagger-test-2)
|
| 79 |
|
| 80 |
+
|Name|Description|
|
| 81 |
+
|-|-|
|
| 82 |
+
|Images count|5000|
|
| 83 |
+
|Supported tags|9517 general tags. Character and rating tags are not included. See all labels in [config.json](config.json)|
|
| 84 |
+
|Image rating|4000 for `general` and 1000 for `sensitive,questionable,explicit`|
|
| 85 |
+
|Copyright tags|`original` only|
|
| 86 |
+
|Image score range (on search)|min:10, max150|
|
| 87 |
|
| 88 |
## Training procedure
|
| 89 |
|
| 90 |
+
- Loss function: AsymmetricLossOptimized ([Asymmetric Loss](https://github.com/Alibaba-MIIL/ASL))
|
| 91 |
+
- `gamma_neg=4, gamma_pos=1, clip=0.05, eps=1e-8, disable_torch_grad_focal_loss=False`
|
| 92 |
+
|
| 93 |
### Training hyperparameters
|
| 94 |
|
| 95 |
The following hyperparameters were used during training:
|