OCR-LayoutLMv3-Invoice

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

Loss: 0.3159
Precision: 0.8765
Recall: 0.8812
F1: 0.8789
Accuracy: 0.9268

Model description

More information needed

Training and evaluation data

More information needed

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: 6000

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	0.16	100	1.5032	0.4934	0.1444	0.2234	0.6064
No log	0.32	200	1.0282	0.5884	0.4420	0.5048	0.7385
No log	0.47	300	0.7856	0.7448	0.6205	0.6770	0.8133
No log	0.63	400	0.6464	0.7736	0.6689	0.7174	0.8399
1.1733	0.79	500	0.5672	0.7609	0.7303	0.7453	0.8557
1.1733	0.95	600	0.5055	0.7658	0.7652	0.7655	0.8677
1.1733	1.1	700	0.4735	0.7946	0.7848	0.7897	0.8784
1.1733	1.26	800	0.4414	0.7962	0.7946	0.7954	0.8818
1.1733	1.42	900	0.4094	0.8176	0.8064	0.8120	0.8894
0.5047	1.58	1000	0.3971	0.8219	0.8248	0.8234	0.8961
0.5047	1.74	1100	0.4082	0.7993	0.8362	0.8174	0.8927
0.5047	1.89	1200	0.3797	0.8240	0.8317	0.8278	0.8962
0.5047	2.05	1300	0.3597	0.8326	0.8331	0.8329	0.9020
0.5047	2.21	1400	0.3544	0.8462	0.8283	0.8371	0.9020
0.368	2.37	1500	0.3374	0.8428	0.8435	0.8432	0.9056
0.368	2.52	1600	0.3364	0.8406	0.8522	0.8464	0.9089
0.368	2.68	1700	0.3404	0.8467	0.8536	0.8501	0.9107
0.368	2.84	1800	0.3319	0.8405	0.8501	0.8453	0.9090
0.368	3.0	1900	0.3324	0.8584	0.8492	0.8538	0.9117
0.2949	3.15	2000	0.3204	0.8691	0.8404	0.8545	0.9119
0.2949	3.31	2100	0.3107	0.8599	0.8547	0.8573	0.9162
0.2949	3.47	2200	0.3169	0.8680	0.8489	0.8584	0.9146
0.2949	3.63	2300	0.3190	0.8683	0.8519	0.8600	0.9152
0.2949	3.79	2400	0.2975	0.8631	0.8617	0.8624	0.9182
0.2438	3.94	2500	0.3040	0.8566	0.8640	0.8603	0.9171
0.2438	4.1	2600	0.3045	0.8585	0.8642	0.8613	0.9181
0.2438	4.26	2700	0.3139	0.8498	0.8748	0.8621	0.9160
0.2438	4.42	2800	0.2985	0.8642	0.8672	0.8657	0.9214
0.2438	4.57	2900	0.3047	0.8688	0.8694	0.8691	0.9214
0.2028	4.73	3000	0.2986	0.8686	0.8695	0.8691	0.9207
0.2028	4.89	3100	0.3135	0.8628	0.8755	0.8691	0.9197
0.2028	5.05	3200	0.2927	0.8656	0.8755	0.8705	0.9217
0.2028	5.21	3300	0.2992	0.8724	0.8697	0.8711	0.9228
0.2028	5.36	3400	0.2975	0.8831	0.8639	0.8734	0.9244
0.1814	5.52	3500	0.2897	0.8736	0.8788	0.8762	0.9250
0.1814	5.68	3600	0.3118	0.8674	0.8751	0.8712	0.9216
0.1814	5.84	3700	0.2974	0.8735	0.8779	0.8757	0.9237
0.1814	5.99	3800	0.2957	0.8696	0.8815	0.8755	0.9240
0.1814	6.15	3900	0.3120	0.8698	0.8817	0.8757	0.9250
0.1602	6.31	4000	0.3080	0.8715	0.8800	0.8757	0.9238
0.1602	6.47	4100	0.3031	0.8767	0.8788	0.8777	0.9261
0.1602	6.62	4200	0.3146	0.8699	0.8784	0.8741	0.9227
0.1602	6.78	4300	0.3085	0.8717	0.8788	0.8752	0.9248
0.1602	6.94	4400	0.3023	0.8749	0.8756	0.8752	0.9250
0.1383	7.1	4500	0.3025	0.8860	0.8735	0.8797	0.9252
0.1383	7.26	4600	0.3026	0.8775	0.8810	0.8792	0.9272
0.1383	7.41	4700	0.3146	0.8715	0.8832	0.8773	0.9251
0.1383	7.57	4800	0.3113	0.8769	0.8803	0.8786	0.9275
0.1383	7.73	4900	0.3073	0.8797	0.8786	0.8792	0.9261
0.1306	7.89	5000	0.3163	0.8714	0.8828	0.8770	0.9248
0.1306	8.04	5100	0.3163	0.8753	0.8810	0.8781	0.9250
0.1306	8.2	5200	0.3132	0.8743	0.8804	0.8773	0.9257
0.1306	8.36	5300	0.3119	0.8735	0.8837	0.8786	0.9264
0.1306	8.52	5400	0.3145	0.8826	0.8779	0.8802	0.9272
0.1174	8.68	5500	0.3166	0.8776	0.8811	0.8794	0.9261
0.1174	8.83	5600	0.3146	0.8776	0.8814	0.8795	0.9260
0.1174	8.99	5700	0.3135	0.8763	0.8826	0.8795	0.9271
0.1174	9.15	5800	0.3154	0.8794	0.8818	0.8806	0.9275
0.1174	9.31	5900	0.3152	0.8788	0.8817	0.8802	0.9274
0.11	9.46	6000	0.3159	0.8765	0.8812	0.8789	0.9268

Framework versions

Transformers 4.25.0.dev0
Pytorch 1.12.1
Datasets 2.6.1
Tokenizers 0.13.1

作者:

Tao Jin

数据集大小:

483.75 MB