ViTFineTuned

This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the KTH-TIPS2-b dataset. It achieves the following results on the evaluation set:

• Loss: 0.0075
• Accuracy: 1.0

Model description

Transfer learning by fine tuning the Vision Transformer by Google on KTP-TIP2-b dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0005
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.2859	0.99	67	0.2180	0.9784
0.293	1.99	134	0.3308	0.9185
0.1444	2.99	201	0.1532	0.9568
0.0833	3.99	268	0.0515	0.9856
0.1007	4.99	335	0.0295	0.9904
0.0372	5.99	402	0.0574	0.9808
0.0919	6.99	469	0.0537	0.9880
0.0135	7.99	536	0.0117	0.9952
0.0472	8.99	603	0.0075	1.0
0.0151	9.99	670	0.0048	1.0
0.0052	10.99	737	0.0073	0.9976
0.0109	11.99	804	0.0198	0.9952
0.0033	12.99	871	0.0066	0.9976
0.011	13.99	938	0.0067	0.9976
0.0032	14.99	1005	0.0060	0.9976

Framework versions

• Transformers 4.20.1
• Pytorch 1.12.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1