metadata

license: cc-by-nc-sa-4.0
base_model: microsoft/layoutlmv2-base-uncased
tags:
  - generated_from_trainer
datasets:
  - cord
metrics:
  - precision
  - recall
  - f1
  - accuracy
model-index:
  - name: layoutlmv2-finetuned-cord
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: cord
          type: cord
          config: cord
          split: validation
          args: cord
        metrics:
          - name: Precision
            type: precision
            value: 0.9652945924132365
          - name: Recall
            type: recall
            value: 0.9676375404530745
          - name: F1
            type: f1
            value: 0.9664646464646465
          - name: Accuracy
            type: accuracy
            value: 0.9702653247941445

overfitting issue

I use this colab: https://colab.research.google.com/drive/1AXh3G3-VmbMWlwbSvesVIurzNlcezTce?usp=sharing

to Fine tuning LayoutLMv2ForTokenClassification on CORD dataset

here is the result: https://huggingface.co/doc2txt/layoutlmv2-finetuned-cord

F1: 0.9665

and indeed the result are pretty amazing when running on the test set, however when running on any other receipt (printed or pdf) the result are completely off

So from some reason the model is overfitting to the cord dataset, even though I use similar images for testing.

I don't think that there is a Data leakage unless the cord DS is not clean (which I assume it is clean)

What could be the reason for this? Is it some inherent property of LayoutLM? The LayoutLM models are somewhat old, and it seems deserted...

I don't have much experience so I would appreciate any info Thanks

here is an example code of how to run this model on a specific img folder: https://huggingface.co/doc2txt/layoutlmv2-finetuned-cord/blob/main/LayoutLMv2Main_cord2_gOcr_folder.py

layoutlmv2-finetuned-cord

This model is a fine-tuned version of microsoft/layoutlmv2-base-uncased on the cord dataset. It achieves the following results on the evaluation set:

Loss: 0.2819
Precision: 0.9653
Recall: 0.9676
F1: 0.9665
Accuracy: 0.9703

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 2
eval_batch_size: 8
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: 5
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	400	1.2752	0.8527	0.8382	0.8454	0.8481
1.9583	2.0	800	0.6372	0.8799	0.8948	0.8873	0.9021
0.7097	3.0	1200	0.4255	0.9241	0.9264	0.9253	0.9414
0.3845	4.0	1600	0.3021	0.9414	0.9482	0.9448	0.9611
0.2699	5.0	2000	0.2819	0.9653	0.9676	0.9665	0.9703

Framework versions

Transformers 4.37.2
Pytorch 2.1.0+cu121
Datasets 2.16.1
Tokenizers 0.15.1