LayoutLM-v3 model fine-tuned on invoice dataset
This model is a fine-tuned version of microsoft/layoutlmv3-base on the invoice dataset.
We use Microsoftβs LayoutLMv3 trained on Invoice Dataset to predict the Biller Name, Biller Address, Biller post_code, Due_date, GST, Invoice_date, Invoice_number, Subtotal and Total. To use it, simply upload an image or use the example image below. Results will show up in a few seconds.
It achieves the following results on the evaluation set:
- Loss: 0.0012
- Precision: 1.0
- Recall: 1.0
- F1: 1.0
- Accuracy: 1.0
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
All the training codes are available from the below GitHub link.
https://github.com/Theivaprakasham/layoutlmv3
The model can be evaluated at the HuggingFace Spaces link:
https://huggingface.co/spaces/Theivaprakasham/layoutlmv3_invoice
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- training_steps: 2000
Training results
Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
---|---|---|---|---|---|---|---|
No log | 2.0 | 100 | 0.0878 | 0.968 | 0.9817 | 0.9748 | 0.9966 |
No log | 4.0 | 200 | 0.0241 | 0.972 | 0.9858 | 0.9789 | 0.9971 |
No log | 6.0 | 300 | 0.0186 | 0.972 | 0.9858 | 0.9789 | 0.9971 |
No log | 8.0 | 400 | 0.0184 | 0.9854 | 0.9574 | 0.9712 | 0.9956 |
0.1308 | 10.0 | 500 | 0.0121 | 0.972 | 0.9858 | 0.9789 | 0.9971 |
0.1308 | 12.0 | 600 | 0.0076 | 0.9939 | 0.9878 | 0.9908 | 0.9987 |
0.1308 | 14.0 | 700 | 0.0047 | 1.0 | 0.9959 | 0.9980 | 0.9996 |
0.1308 | 16.0 | 800 | 0.0036 | 0.9960 | 0.9980 | 0.9970 | 0.9996 |
0.1308 | 18.0 | 900 | 0.0045 | 0.9960 | 0.9980 | 0.9970 | 0.9996 |
0.0069 | 20.0 | 1000 | 0.0043 | 0.9960 | 0.9980 | 0.9970 | 0.9996 |
0.0069 | 22.0 | 1100 | 0.0016 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0069 | 24.0 | 1200 | 0.0015 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0069 | 26.0 | 1300 | 0.0014 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0069 | 28.0 | 1400 | 0.0013 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0026 | 30.0 | 1500 | 0.0012 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0026 | 32.0 | 1600 | 0.0012 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0026 | 34.0 | 1700 | 0.0011 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0026 | 36.0 | 1800 | 0.0011 | 1.0 | 1.0 | 1.0 | 1.0 |
0.0026 | 38.0 | 1900 | 0.0011 | 1.0 | 1.0 | 1.0 | 1.0 |
0.002 | 40.0 | 2000 | 0.0011 | 1.0 | 1.0 | 1.0 | 1.0 |
Framework versions
- Transformers 4.20.0.dev0
- Pytorch 1.11.0+cu113
- Datasets 2.2.2
- Tokenizers 0.12.1
- Downloads last month
- 1,007
Spaces using Theivaprakasham/layoutlmv3-finetuned-invoice 11
Evaluation results
- Precision on Invoiceself-reported1.000
- Recall on Invoiceself-reported1.000
- F1 on Invoiceself-reported1.000
- Accuracy on Invoiceself-reported1.000