Update README.md
Browse files
README.md
CHANGED
|
@@ -1,86 +1,9 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
---
|
| 4 |
|
| 5 |
# LayoutReader
|
| 6 |
|
| 7 |
-
|
| 8 |
-
1. upload models to huggingface
|
| 9 |
-
2. explain why this repo
|
| 10 |
-
3. explain the new dataset
|
| 11 |
-
4. build docker image
|
| 12 |
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
### Build Dataset
|
| 16 |
-
|
| 17 |
-
```bash
|
| 18 |
-
python tools.py cache-dataset-spans --help
|
| 19 |
-
```
|
| 20 |
-
|
| 21 |
-
### Train
|
| 22 |
-
|
| 23 |
-
```bash
|
| 24 |
-
bash train.sh
|
| 25 |
-
```
|
| 26 |
-
|
| 27 |
-
### Eval
|
| 28 |
-
|
| 29 |
-
```bash
|
| 30 |
-
python eval.py --help
|
| 31 |
-
```
|
| 32 |
-
|
| 33 |
-
## Spans-Level Results
|
| 34 |
-
|
| 35 |
-
One bbox contains multiple tokens. Usually, parse pdf file to get bbox. Training data is generated by `tools.py`.
|
| 36 |
-
|
| 37 |
-
> only use the first part of test file
|
| 38 |
-
|
| 39 |
-
| Method | shuf | BLEU Idx | BLEU Token |
|
| 40 |
-
|----------------------------|------|----------|------------|
|
| 41 |
-
| Heuristic Method | no | 44.4 | 70.7 |
|
| 42 |
-
| LayoutReader (layout only) | no | 95.3 | 97.8 |
|
| 43 |
-
| LayoutReader (layout only) | yes | 95.0 | 97.6 |
|
| 44 |
-
|
| 45 |
-
## Tokens-Level Results
|
| 46 |
-
|
| 47 |
-
One bbox contains only one token.
|
| 48 |
-
|
| 49 |
-
### New eval script
|
| 50 |
-
|
| 51 |
-
> only use the first part of test file
|
| 52 |
-
|
| 53 |
-
| Method | shuf | BLEU Idx | BLEU Token |
|
| 54 |
-
|-----------------------------|------|----------|------------|
|
| 55 |
-
| Heuristic Method | no | 78.3 | 79.4 |
|
| 56 |
-
| LayoutReader (layout only) | no | 98.0 | 98.2 |
|
| 57 |
-
| LayoutReader (layout only) | yes | 97.8 | 98.0 |
|
| 58 |
-
| LayoutReader (public model) | no | 98.0 | 98.3 |
|
| 59 |
-
|
| 60 |
-
### Old eval script (from original paper)
|
| 61 |
-
|
| 62 |
-
* Evaluation results of the LayoutReader on the reading order detection task, where the source-side of training/testing
|
| 63 |
-
data is in the left-to-right and top-to-bottom order
|
| 64 |
-
|
| 65 |
-
| Method | Encoder | BLEU | ARD |
|
| 66 |
-
|----------------------------|------------------------|--------|------|
|
| 67 |
-
| Heuristic Method | - | 0.6972 | 8.46 |
|
| 68 |
-
| LayoutReader (layout only) | LayoutLM (layout only) | 0.9732 | 2.31 |
|
| 69 |
-
| LayoutReader | LayoutLM | 0.9819 | 1.75 |
|
| 70 |
-
|
| 71 |
-
* Input order study with left-to-right and top-to-bottom inputs in evaluation, where r is the proportion of
|
| 72 |
-
shuffled samples in training.
|
| 73 |
-
|
| 74 |
-
| Method | BLEU | BLEU | BLEU | ARD | ARD | ARD |
|
| 75 |
-
|----------------------------|--------|--------|--------|--------|-------|------|
|
| 76 |
-
| | r=100% | r=50% | r=0% | r=100% | r=50% | r=0% |
|
| 77 |
-
| LayoutReader (layout only) | 0.9701 | 0.9729 | 0.9732 | 2.85 | 2.61 | 2.31 |
|
| 78 |
-
| LayoutReader | 0.9765 | 0.9788 | 0.9819 | 2.50 | 2.24 | 1.75 |
|
| 79 |
-
|
| 80 |
-
* Input order study with token-shuffled inputs in evaluation, where r is the proportion of shuffled samples in training.
|
| 81 |
-
|
| 82 |
-
| Method | BLEU | BLEU | BLEU | ARD | ARD | ARD |
|
| 83 |
-
|----------------------------|--------|--------|--------|--------|-------|--------|
|
| 84 |
-
| | r=100% | r=50% | r=0% | r=100% | r=50% | r=0% |
|
| 85 |
-
| LayoutReader (layout only) | 0.9718 | 0.9714 | 0.1331 | 2.72 | 2.82 | 105.40 |
|
| 86 |
-
| LayoutReader | 0.9772 | 0.9770 | 0.1783 | 2.48 | 2.46 | 72.94 |
|
|
|
|
| 1 |
---
|
| 2 |
+
library_name: transformers
|
| 3 |
---
|
| 4 |
|
| 5 |
# LayoutReader
|
| 6 |
|
| 7 |
+
A reading order prediction model. Turn bboxes extracted from PDF or detected by OCR into readable order.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
+
Please refer to [Github](https://github.com/ppaanngggg/layoutreader) for more details.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|