Update README.md
Browse files
README.md
CHANGED
|
@@ -5,6 +5,8 @@ tags:
|
|
| 5 |
- classification
|
| 6 |
base_model:
|
| 7 |
- google/vit-base-patch16-224
|
|
|
|
|
|
|
| 8 |
pipeline_tag: image-classification
|
| 9 |
license: mit
|
| 10 |
---
|
|
@@ -21,63 +23,50 @@ HF 😊 hub support for the model
|
|
| 21 |
## Versions 🏁
|
| 22 |
|
| 23 |
There are currently 2 version of the model available for download, both of them have the same set of categories,
|
| 24 |
-
but different data annotations. The latest `v2.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
|
| 26 |
-
| Version | Pages | N-page files | PDFs | Description |
|
| 27 |
-
|--------:|:-----:|:------------:|:--------:|:--------------------------------------------------------------|
|
| 28 |
-
| `v1.0` | 10073 | **~104** | **3896** | annotations with mistakes, more heterogenous data |
|
| 29 |
-
| `v1.0` | 11940 | **~509** | **5002** | more diverse pages in each category, less annotation mistakes |
|
| 30 |
|
| 31 |
## Model description 📇
|
| 32 |
|
| 33 |
🔲 Fine-tuned model repository: vit-historical-page [^1] 🔗
|
| 34 |
|
| 35 |
-
🔳 Base model repository:
|
| 36 |
|
| 37 |
### Data 📜
|
| 38 |
|
| 39 |
-
Training set of the model: **8950** images for
|
|
|
|
|
|
|
| 40 |
|
| 41 |
-
Training set of the model: **
|
| 42 |
|
| 43 |
### Categories 🏷️
|
| 44 |
|
| 45 |
-
**v1.0 version Categories 🪧**:
|
| 46 |
-
|
| 47 |
-
| Label️ | Ratio | Description |
|
| 48 |
-
|----------:|:------:|:------------------------------------------------------------------------------|
|
| 49 |
-
| `DRAW` | 11.89% | **📈 - drawings, maps, paintings with text** |
|
| 50 |
-
| `DRAW_L` | 8.17% | **📈📏 - drawings, etc with a table legend or inside tabular layout / forms** |
|
| 51 |
-
| `LINE_HW` | 5.99% | **✏️📏 - handwritten text lines inside tabular layout / forms** |
|
| 52 |
-
| `LINE_P` | 6.06% | **📏 - printed text lines inside tabular layout / forms** |
|
| 53 |
-
| `LINE_T` | 13.39% | **📏 - machine typed text lines inside tabular layout / forms** |
|
| 54 |
-
| `PHOTO` | 10.21% | **🌄 - photos with text** |
|
| 55 |
-
| `PHOTO_L` | 7.86% | **🌄📏 - photos inside tabular layout / forms or with a tabular annotation** |
|
| 56 |
-
| `TEXT` | 8.58% | **📰 - mixed types of printed and handwritten texts** |
|
| 57 |
-
| `TEXT_HW` | 7.36% | **✏️📄 - only handwritten text** |
|
| 58 |
-
| `TEXT_P` | 6.95% | **📄 - only printed text** |
|
| 59 |
-
| `TEXT_T` | 13.53% | **📄 - only machine typed text** |
|
| 60 |
-
|
| 61 |
-
**v2.0 version Categories 🪧**:
|
| 62 |
-
|
| 63 |
-
| Label️ | Ratio | Description |
|
| 64 |
-
|----------:|:-----:|:------------------------------------------------------------------------------|
|
| 65 |
-
| `DRAW` | 9.12% | **📈 - drawings, maps, paintings with text** |
|
| 66 |
-
| `DRAW_L` | 9.14% | **📈📏 - drawings, etc with a table legend or inside tabular layout / forms** |
|
| 67 |
-
| `LINE_HW` | 8.84% | **✏️📏 - handwritten text lines inside tabular layout / forms** |
|
| 68 |
-
| `LINE_P` | 9.15% | **📏 - printed text lines inside tabular layout / forms** |
|
| 69 |
-
| `LINE_T` | 9.2% | **📏 - machine typed text lines inside tabular layout / forms** |
|
| 70 |
-
| `PHOTO` | 9.05% | **🌄 - photos with text** |
|
| 71 |
-
| `PHOTO_L` | 9.1% | **🌄📏 - photos inside tabular layout / forms or with a tabular annotation** |
|
| 72 |
-
| `TEXT` | 9.14% | **📰 - mixed types of printed and handwritten texts** |
|
| 73 |
-
| `TEXT_HW` | 9.14% | **✏️📄 - only handwritten text** |
|
| 74 |
-
| `TEXT_P` | 9.07% | **📄 - only printed text** |
|
| 75 |
-
| `TEXT_T` | 9.05% | **📄 - only machine typed text** |
|
| 76 |
-
|
| 77 |
-
Evaluation set (same proportions): **995** images for v1.0
|
| 78 |
-
|
| 79 |
-
Evaluation set (same proportions): **1194** images for v2.0
|
| 80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
|
| 82 |
#### Data preprocessing
|
| 83 |
|
|
@@ -105,31 +94,43 @@ During training the following transforms were applied randomly with a 50% chance
|
|
| 105 |
|
| 106 |
### Results 📊
|
| 107 |
|
| 108 |
-
**
|
|
|
|
|
|
|
|
|
|
|
|
|
| 109 |
|
| 110 |
-
|
| 70 |
|
| 71 |
#### Data preprocessing
|
| 72 |
|
|
|
|
| 94 |
|
| 95 |
### Results 📊
|
| 96 |
|
| 97 |
+
**v2.0** Evaluation set's accuracy (**Top-3**): **95.58%**
|
| 98 |
+
|
| 99 |
+
|
| 100 |
+
|
| 101 |
+
**v2.1** Evaluation set's accuracy (**Top-3**): **99.84%**
|
| 102 |
|
| 103 |
+
|
| 104 |
|
| 105 |
+
**v2.2** Evaluation set's accuracy (**Top-3**): **100.00%**
|
| 106 |
|
| 107 |
+
|
| 108 |
|
| 109 |
+
**v2.0** Evaluation set's accuracy (**Top-1**): **84.96%**
|
| 110 |
|
| 111 |
+
|
| 112 |
|
| 113 |
+
**v2.1** Evaluation set's accuracy (**Top-1**): **96.36%**
|
| 114 |
|
| 115 |
+
|
| 116 |
+
|
| 117 |
+
**v2.2** Evaluation set's accuracy (**Top-1**): **99.61%**
|
| 118 |
+
|
| 119 |
+
|
| 120 |
|
| 121 |
#### Result tables
|
| 122 |
|
| 123 |
+
- **v2.0** Manually ✍ **checked** evaluation dataset results (TOP-3): [model_TOP-3_EVAL.csv](https://github.com/ufal/atrium-page-classification/blob/main/result/tables/20250526-1142_model_v20_TOP-3_EVAL.csv) 🔗
|
| 124 |
+
|
| 125 |
+
- **v2.0** Manually ✍ **checked** evaluation dataset results (TOP-1): [model_TOP-1_EVAL.csv](https://github.com/ufal/atrium-page-classification/blob/main/result/tables/20250526-1148_model_v20_TOP-1_EVAL.csv) 🔗
|
| 126 |
+
|
| 127 |
+
- **v2.1** Manually ✍ **checked** evaluation dataset results (TOP-3): [model_TOP-3_EVAL.csv](https://github.com/ufal/atrium-page-classification/blob/main/result/tables/20250526-1153_model_v21_TOP-3_EVAL.csv) 🔗
|
| 128 |
|
| 129 |
+
- **v2.1** Manually ✍ **checked** evaluation dataset results (TOP-1): [model_TOP-1_EVAL.csv](https://github.com/ufal/atrium-page-classification/blob/main/result/tables/20250526-1151_model_v21_TOP-1_EVAL.csv) 🔗
|
| 130 |
|
| 131 |
+
- **v2.2** Manually ✍ **checked** evaluation dataset results (TOP-3): [model_TOP-3_EVAL.csv](https://github.com/ufal/atrium-page-classification/blob/main/result/tables/20250526-1156_model_v22_TOP-3_EVAL.csv) 🔗
|
| 132 |
|
| 133 |
+
- **v2.2** Manually ✍ **checked** evaluation dataset results (TOP-1): [model_TOP-1_EVAL.csv](https://github.com/ufal/atrium-page-classification/blob/main/result/tables/20250526-1158_model_v22_TOP-1_EVAL.csv) 🔗
|
| 134 |
|
| 135 |
#### Table columns
|
| 136 |
|
|
|
|
| 151 |
- **Developed by** UFAL [^5] 👥
|
| 152 |
- **Funded by** ATRIUM [^4] 💰
|
| 153 |
- **Shared by** ATRIUM [^4] & UFAL [^5]
|
| 154 |
+
- **Model type:** fine-tuned ViT with a 224x224 [^2] 🔗 or 384x384 [^13] [^14] 🔗 resolution size
|
| 155 |
|
| 156 |
**©️ 2022 UFAL & ATRIUM**
|
| 157 |
|
|
|
|
| 160 |
[^3]: https://github.com/ufal/atrium-page-classification
|
| 161 |
[^4]: https://atrium-research.eu/
|
| 162 |
[^5]: https://ufal.mff.cuni.cz/home-page
|
| 163 |
+
[^6]: https://huggingface.co/google/vit-base-patch16-384
|
| 164 |
+
[^7]: https://huggingface.co/google/vit-large-patch16-384
|