PawanKrGunjan/license_plate_recognizer

anpr-small

This model is a fine-tuned version of microsoft/trocr-small-printed on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3888
• Cer: 0.0310

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.8237	1.0	128	0.5120	0.0769
0.3165	2.0	256	0.4186	0.0652
0.2877	3.0	384	0.3774	0.0578
0.2555	4.0	512	0.4401	0.0610
0.1626	5.0	640	0.3647	0.0461
0.1831	6.0	768	0.3761	0.0486
0.1247	7.0	896	0.3560	0.0491
0.1276	8.0	1024	0.3484	0.0464
0.0957	9.0	1152	0.3934	0.0528
0.0931	10.0	1280	0.3324	0.0387
0.0928	11.0	1408	0.3758	0.0430
0.0786	12.0	1536	0.3863	0.0437
0.0622	13.0	1664	0.3677	0.0405
0.0505	14.0	1792	0.3927	0.0471
0.042	15.0	1920	0.3978	0.0442
0.0435	16.0	2048	0.3611	0.0364
0.0288	17.0	2176	0.3902	0.0416
0.0302	18.0	2304	0.3711	0.0380
0.0291	19.0	2432	0.3954	0.0380
0.0151	20.0	2560	0.3875	0.0376
0.0199	21.0	2688	0.3894	0.0364
0.0221	22.0	2816	0.4208	0.0400
0.0088	23.0	2944	0.3960	0.0345
0.0111	24.0	3072	0.3765	0.0338
0.0061	25.0	3200	0.3954	0.0336
0.0081	26.0	3328	0.3880	0.0329
0.0074	27.0	3456	0.3827	0.0313
0.0064	28.0	3584	0.3888	0.0310
0.0096	29.0	3712	0.3852	0.0322
0.0036	30.0	3840	0.3854	0.0317

Framework versions

• Transformers 4.44.2
• Pytorch 2.6.0+cu124
• Datasets 3.0.1
• Tokenizers 0.19.1