Document Understanding model (finetuned LayoutXLM base at paragraph level on DocLayNet base)

This model is a fine-tuned version of microsoft/layoutxlm-base with the DocLayNet base dataset. It achieves the following results on the evaluation set:

Loss: 0.1796
Precision: 0.8062
Recall: 0.7441
F1: 0.7739
Token Accuracy: 0.9693
Paragraph Accuracy: 0.8655

Accuracy at paragraph level

Paragraph Accuracy: 86.55%
Accuracy by label
- Caption: 63.76%
- Footnote: 31.91%
- Formula: 95.33%
- List-item: 79.31%
- Page-footer: 99.51%
- Page-header: 88.75%
- Picture: 90.91%
- Section-header: 83.16%
- Table: 68.25%
- Text: 91.37%
- Title: 50.0%

References

Blog posts

Layout XLM base
LiLT base - (02/16/2023) Document AI | Inference APP and fine-tuning notebook for Document Understanding at paragraph level - (02/14/2023) Document AI | Inference APP for Document Understanding at line level - (02/10/2023) Document AI | Document Understanding model at line level with LiLT, Tesseract and DocLayNet dataset - (01/31/2023) Document AI | DocLayNet image viewer APP - (01/27/2023) Document AI | Processing of DocLayNet dataset to be used by layout models of the Hugging Face hub (finetuning, inference)

Notebooks (paragraph level)

Notebooks (line level)

APP

You can test this model with this APP in Hugging Face Spaces: Inference APP for Document Understanding at paragraph level (v2).

You can run as well the corresponding notebook: Document AI | Inference APP at paragraph level with a Document Understanding model (LayoutXLM base fine-tuned on DocLayNet dataset)

DocLayNet dataset

DocLayNet dataset (IBM) provides page-by-page layout segmentation ground-truth using bounding-boxes for 11 distinct class labels on 80863 unique pages from 6 document categories.

Until today, the dataset can be downloaded through direct links or as a dataset from Hugging Face datasets:

direct links: doclaynet_core.zip (28 GiB), doclaynet_extra.zip (7.5 GiB)
Hugging Face dataset library: dataset DocLayNet

Paper: DocLayNet: A Large Human-Annotated Dataset for Document-Layout Analysis (06/02/2022)

Model description

The model was finetuned at paragraph level on chunk of 512 tokens with overlap of 128 tokens. Thus, the model was trained with all layout and text data of all pages of the dataset.

At inference time, a calculation of best probabilities give the label to each paragraph bounding boxes.

Inference

See notebook: Document AI | Inference at paragraph level with a Document Understanding model (LayoutXLM base fine-tuned on DocLayNet dataset)

Training and evaluation data

See notebook: Document AI | Fine-tune LayoutXLM base on DocLayNet base in any language at paragraph level (chunk of 512 tokens with overlap)

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 4
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Accuracy	F1	Validation Loss	Precision	Recall
No log	0.11	200	0.8842	0.1066	0.4428	0.1154	0.0991
No log	0.21	400	0.9243	0.4440	0.3040	0.4548	0.4336
0.7241	0.32	600	0.9359	0.5544	0.2265	0.5330	0.5775
0.7241	0.43	800	0.9479	0.6015	0.2140	0.6013	0.6017
0.2343	0.53	1000	0.9402	0.6132	0.2852	0.6642	0.5695
0.2343	0.64	1200	0.9540	0.6604	0.1694	0.6565	0.6644
0.2343	0.75	1400	0.9354	0.6198	0.2308	0.5119	0.7854
0.1913	0.85	1600	0.9594	0.6590	0.1601	0.7190	0.6082
0.1913	0.96	1800	0.9541	0.6597	0.1671	0.5790	0.7664
0.1346	1.07	2000	0.9612	0.6986	0.1580	0.6838	0.7140
0.1346	1.17	2200	0.9597	0.6897	0.1423	0.6618	0.7200
0.1346	1.28	2400	0.9663	0.6980	0.1580	0.7490	0.6535
0.098	1.39	2600	0.9616	0.6800	0.1394	0.7044	0.6573
0.098	1.49	2800	0.9686	0.7251	0.1756	0.6893	0.7649
0.0999	1.6	3000	0.9636	0.6985	0.1542	0.7127	0.6848
0.0999	1.71	3200	0.9670	0.7097	0.1187	0.7538	0.6705
0.0999	1.81	3400	0.9585	0.7427	0.1793	0.7602	0.7260
0.0972	1.92	3600	0.9621	0.7189	0.1836	0.7576	0.6839
0.0972	2.03	3800	0.9642	0.7189	0.1465	0.7388	0.6999
0.0662	2.13	4000	0.9691	0.7450	0.1409	0.7615	0.7292
0.0662	2.24	4200	0.9615	0.7432	0.1720	0.7435	0.7429
0.0662	2.35	4400	0.9667	0.7338	0.1440	0.7469	0.7212
0.0581	2.45	4600	0.9657	0.7135	0.1928	0.7458	0.6839
0.0581	2.56	4800	0.9692	0.7378	0.1645	0.7467	0.7292
0.0538	2.67	5000	0.9656	0.7619	0.1517	0.7700	0.7541
0.0538	2.77	5200	0.9684	0.7728	0.1676	0.8227	0.7286
0.0538	2.88	5400	0.9725	0.7608	0.1277	0.7865	0.7367
0.0432	2.99	5600	0.9693	0.7784	0.1532	0.7891	0.7681
0.0432	3.09	5800	0.9692	0.7783	0.1701	0.8067	0.7519
0.0272	3.2	6000	0.9732	0.7798	0.1159	0.8072	0.7542
0.0272	3.3	6200	0.9720	0.7797	0.1835	0.7926	0.7672
0.0272	3.41	6400	0.9730	0.7894	0.1481	0.8183	0.7624
0.0274	3.52	6600	0.9686	0.7655	0.1552	0.7958	0.7373
0.0274	3.62	6800	0.9698	0.7724	0.1523	0.8068	0.7407
0.0246	3.73	7000	0.9691	0.7720	0.1673	0.7960	0.7493
0.0246	3.84	7200	0.9688	0.7695	0.1333	0.7986	0.7424
0.0246	3.94	7400	0.1796	0.8062	0.7441	0.7739	0.9693

Framework versions

Transformers 4.27.3
Pytorch 1.10.0+cu111
Datasets 2.10.1
Tokenizers 0.13.2

Other models

Line level
- Document Understanding model (finetuned LiLT base at line level on DocLayNet base) (accuracy | tokens: 85.84% - lines: 91.97%)
- Document Understanding model (finetuned LayoutXLM base at line level on DocLayNet base) (accuracy | tokens: 93.73% - lines: ...)
Paragraph level
- Document Understanding model (finetuned LiLT base at paragraph level on DocLayNet base) (accuracy | tokens: 86.34% - paragraphs: 68.15%)
- Document Understanding model (finetuned LayoutXLM base at paragraph level on DocLayNet base) (accuracy | tokens: 96.93% - paragraphs: 86.55%)

pierreguillou
/

layout-xlm-base-finetuned-with-DocLayNet-base-at-paragraphlevel-ml512