Hugging Face's logo

jinhybr
/
OCR-LayoutLMv3

Token Classification
Transformers
PyTorch
TensorBoard
layoutlmv3
Generated from Trainer
Eval Results
Inference Endpoints
Model card Files Metrics Community
2
OCR-LayoutLMv3 / README.md
librarian-bot's picture
librarian-bot
Librarian Bot: Add base_model information to model
8e3dec0
preview code
|
raw
history blame
4.33 kB
metadata
license: cc-by-nc-sa-4.0
tags:
  - generated_from_trainer
datasets:
  - funsd-layoutlmv3
metrics:
  - precision
  - recall
  - f1
  - accuracy
base_model: microsoft/layoutlmv3-base
model-index:
  - name: OCR-LayoutLMv3
    results:
      - task:
          type: token-classification
          name: Token Classification
        dataset:
          name: funsd-layoutlmv3
          type: funsd-layoutlmv3
          config: funsd
          split: train
          args: funsd
        metrics:
          - type: precision
            value: 0.8988653182042428
            name: Precision
          - type: recall
            value: 0.905116741182315
            name: Recall
          - type: f1
            value: 0.9019801980198019
            name: F1
          - type: accuracy
            value: 0.8403661000832046
            name: Accuracy

OCR-LayoutLMv3

This model is a fine-tuned version of microsoft/layoutlmv3-base on the funsd-layoutlmv3 dataset. It achieves the following results on the evaluation set:

  • Loss: 0.9788
  • Precision: 0.8989
  • Recall: 0.9051
  • F1: 0.9020
  • Accuracy: 0.8404

Model description

LayoutLMv3 is a pre-trained multimodal Transformer for Document AI with unified text and image masking. The simple unified architecture and training objectives make LayoutLMv3 a general-purpose pre-trained model. For example, LayoutLMv3 can be fine-tuned for both text-centric tasks, including form understanding, receipt understanding, and document visual question answering, and image-centric tasks such as document image classification and document layout analysis.

LayoutLMv3: Pre-training for Document AI with Unified Text and Image Masking Yupan Huang, Tengchao Lv, Lei Cui, Yutong Lu, Furu Wei, Preprint 2022.

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 1.33 100 0.6966 0.7418 0.8063 0.7727 0.7801
No log 2.67 200 0.5767 0.8104 0.8644 0.8365 0.8117
No log 4.0 300 0.5355 0.8246 0.8852 0.8539 0.8295
No log 5.33 400 0.5240 0.8706 0.8922 0.8813 0.8427
0.5326 6.67 500 0.6337 0.8528 0.8778 0.8651 0.8260
0.5326 8.0 600 0.6870 0.8698 0.8828 0.8762 0.8240
0.5326 9.33 700 0.6584 0.8723 0.9061 0.8889 0.8342
0.5326 10.67 800 0.7186 0.8868 0.9031 0.8949 0.8335
0.5326 12.0 900 0.6822 0.9040 0.9076 0.9058 0.8526
0.1248 13.33 1000 0.7042 0.8872 0.9021 0.8946 0.8511
0.1248 14.67 1100 0.7920 0.9027 0.9036 0.9032 0.8480
0.1248 16.0 1200 0.8052 0.8964 0.9151 0.9056 0.8389
0.1248 17.33 1300 0.8932 0.8995 0.9066 0.9030 0.8329
0.1248 18.67 1400 0.8728 0.8950 0.9061 0.9005 0.8398
0.0442 20.0 1500 0.9051 0.8960 0.9116 0.9037 0.8347
0.0442 21.33 1600 0.9587 0.8947 0.9031 0.8989 0.8401
0.0442 22.67 1700 0.9822 0.9042 0.9046 0.9044 0.8389
0.0442 24.0 1800 0.9734 0.9043 0.9061 0.9052 0.8391
0.0442 25.33 1900 0.9842 0.9042 0.9091 0.9066 0.8410
0.0225 26.67 2000 0.9788 0.8989 0.9051 0.9020 0.8404

Framework versions

  • Transformers 4.25.0.dev0
  • Pytorch 1.12.1
  • Datasets 2.6.1
  • Tokenizers 0.13.1