Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,164 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
|
| 4 |
+
tags:
|
| 5 |
+
- onnx
|
| 6 |
+
- document-understanding
|
| 7 |
+
- layout-detection
|
| 8 |
+
- table-detection
|
| 9 |
+
- faria
|
| 10 |
+
|
| 11 |
+
pipeline_tag: object-detection
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# Faria ONNX Models
|
| 15 |
+
|
| 16 |
+
Pre-exported ONNX models used by [Faria](https://github.com/exto360-inc/faria), a document processing library with ML-powered
|
| 17 |
+
layout detection and table extraction. These files are ready for direct use with ONNX Runtime — no Python or conversion step
|
| 18 |
+
required.
|
| 19 |
+
|
| 20 |
+
## Models
|
| 21 |
+
|
| 22 |
+
### `detr_layout_detection.onnx` (~350 MB)
|
| 23 |
+
|
| 24 |
+
Document layout detection. Identifies structural elements across a page.
|
| 25 |
+
|
| 26 |
+
- **Source:** [`cmarkea/detr-layout-detection`](https://huggingface.co/cmarkea/detr-layout-detection)
|
| 27 |
+
- **ONNX opset:** 14
|
| 28 |
+
|
| 29 |
+
**Input**
|
| 30 |
+
|
| 31 |
+
| Name | Shape | Type |
|
| 32 |
+
|----------------|------------------------|---------|
|
| 33 |
+
| `pixel_values` | `[batch, 3, 800, 800]` | float32 |
|
| 34 |
+
|
| 35 |
+
**Outputs**
|
| 36 |
+
|
| 37 |
+
| Name | Shape | Type | Description |
|
| 38 |
+
|--------------|--------------------|---------|--------------------------------------------|
|
| 39 |
+
| `logits` | `[batch, 100, 12]` | float32 | Class scores (11 classes + no-object) |
|
| 40 |
+
| `pred_boxes` | `[batch, 100, 4]` | float32 | Normalized boxes `(cx, cy, w, h)` |
|
| 41 |
+
|
| 42 |
+
**Class labels (DocLayNet)**
|
| 43 |
+
|
| 44 |
+
| Index | Label |
|
| 45 |
+
|-------|----------------|
|
| 46 |
+
| 0 | Caption |
|
| 47 |
+
| 1 | Footnote |
|
| 48 |
+
| 2 | Formula |
|
| 49 |
+
| 3 | List-item |
|
| 50 |
+
| 4 | Page-footer |
|
| 51 |
+
| 5 | Page-header |
|
| 52 |
+
| 6 | Picture |
|
| 53 |
+
| 7 | Section-header |
|
| 54 |
+
| 8 | Table |
|
| 55 |
+
| 9 | Text |
|
| 56 |
+
| 10 | Title |
|
| 57 |
+
| 11 | (no object) |
|
| 58 |
+
|
| 59 |
+
**Post-processing**
|
| 60 |
+
1. Apply softmax to `logits`
|
| 61 |
+
2. Filter by confidence threshold
|
| 62 |
+
3. Convert `(cx, cy, w, h)` → `(x1, y1, x2, y2)`
|
| 63 |
+
4. Scale boxes to image size
|
| 64 |
+
|
| 65 |
+
---
|
| 66 |
+
|
| 67 |
+
### `nemotron_table_structure.onnx` (~200 MB)
|
| 68 |
+
|
| 69 |
+
Table structure recognition.
|
| 70 |
+
|
| 71 |
+
- **Source:** [`nvidia/nemotron-table-structure-v1`](https://huggingface.co/nvidia/nemotron-table-structure-v1)
|
| 72 |
+
- **ONNX opset:** 18
|
| 73 |
+
|
| 74 |
+
**Inputs**
|
| 75 |
+
|
| 76 |
+
| Name | Shape | Type | Description |
|
| 77 |
+
|--------------|---------------------|---------|----------------------------------|
|
| 78 |
+
| `input` | `[1, 3, 1024, 1024]`| float32 | RGB image |
|
| 79 |
+
| `orig_sizes` | `[1, 2]` | int64 | `[height, width]` |
|
| 80 |
+
|
| 81 |
+
**Outputs**
|
| 82 |
+
|
| 83 |
+
| Name | Shape | Type |
|
| 84 |
+
|---------|----------|---------|
|
| 85 |
+
| labels | `[N]` | float32 |
|
| 86 |
+
| boxes | `[N, 4]` | float32 |
|
| 87 |
+
| scores | `[N]` | float32 |
|
| 88 |
+
|
| 89 |
+
**Class labels**
|
| 90 |
+
|
| 91 |
+
| Index | Label |
|
| 92 |
+
|-------|--------|
|
| 93 |
+
| 1 | cell |
|
| 94 |
+
| 2 | row |
|
| 95 |
+
| 3 | column |
|
| 96 |
+
| 4 | header |
|
| 97 |
+
|
| 98 |
+
---
|
| 99 |
+
|
| 100 |
+
## Installation
|
| 101 |
+
|
| 102 |
+
```bash
|
| 103 |
+
curl -fsSL https://raw.githubusercontent.com/exto360-inc/faria-install/main/install.sh | bash -s -- --features idp
|
| 104 |
+
```
|
| 105 |
+
|
| 106 |
+
Or download manually:
|
| 107 |
+
|
| 108 |
+
```bash
|
| 109 |
+
# Layout detection
|
| 110 |
+
curl -fsSL https://huggingface.co/pavan-synkrato360/faria-models/resolve/main/detr_layout_detection.onnx -o detr_layout_detection.onnx
|
| 111 |
+
|
| 112 |
+
# Table structure
|
| 113 |
+
curl -fsSL https://huggingface.co/pavan-synkrato360/faria-models/resolve/main/nemotron_table_structure.onnx -o nemotron_table_structure.onnx
|
| 114 |
+
```
|
| 115 |
+
|
| 116 |
+
---
|
| 117 |
+
|
| 118 |
+
## config.json
|
| 119 |
+
|
| 120 |
+
```json
|
| 121 |
+
{
|
| 122 |
+
"models": {
|
| 123 |
+
"detr_layout_detection": {
|
| 124 |
+
"filename": "detr_layout_detection.onnx",
|
| 125 |
+
"task": "document-layout-detection",
|
| 126 |
+
"source": "cmarkea/detr-layout-detection",
|
| 127 |
+
"onnx_opset": 14,
|
| 128 |
+
"input": {
|
| 129 |
+
"pixel_values": [1, 3, 800, 800]
|
| 130 |
+
},
|
| 131 |
+
"outputs": {
|
| 132 |
+
"logits": [1, 100, 12],
|
| 133 |
+
"pred_boxes": [1, 100, 4]
|
| 134 |
+
},
|
| 135 |
+
"classes": [
|
| 136 |
+
"Caption", "Footnote", "Formula", "List-item",
|
| 137 |
+
"Page-footer", "Page-header", "Picture", "Section-header",
|
| 138 |
+
"Table", "Text", "Title"
|
| 139 |
+
]
|
| 140 |
+
},
|
| 141 |
+
"nemotron_table_structure": {
|
| 142 |
+
"filename": "nemotron_table_structure.onnx",
|
| 143 |
+
"task": "table-structure-recognition",
|
| 144 |
+
"source": "nvidia/nemotron-table-structure-v1",
|
| 145 |
+
"onnx_opset": 18,
|
| 146 |
+
"inputs": {
|
| 147 |
+
"input": [1, 3, 1024, 1024],
|
| 148 |
+
"orig_sizes": [1, 2]
|
| 149 |
+
},
|
| 150 |
+
"outputs": {
|
| 151 |
+
"labels": ["N"],
|
| 152 |
+
"boxes": ["N", 4],
|
| 153 |
+
"scores": ["N"]
|
| 154 |
+
},
|
| 155 |
+
"classes": {
|
| 156 |
+
"1": "cell",
|
| 157 |
+
"2": "row",
|
| 158 |
+
"3": "column",
|
| 159 |
+
"4": "header"
|
| 160 |
+
}
|
| 161 |
+
}
|
| 162 |
+
}
|
| 163 |
+
}
|
| 164 |
+
```
|