vit-emotion-classifier

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

• Loss: 1.3506
• Accuracy: 0.525

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: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	20	2.0656	0.1938
No log	2.0	40	2.0408	0.2625
No log	3.0	60	1.9845	0.275
No log	4.0	80	1.8774	0.35
1.9717	5.0	100	1.7409	0.45
1.9717	6.0	120	1.6349	0.4437
1.9717	7.0	140	1.5541	0.4437
1.9717	8.0	160	1.5007	0.5188
1.9717	9.0	180	1.4531	0.525
1.4968	10.0	200	1.4263	0.5312
1.4968	11.0	220	1.3975	0.5188
1.4968	12.0	240	1.3915	0.525
1.4968	13.0	260	1.3270	0.5375
1.4968	14.0	280	1.3360	0.575
1.2146	15.0	300	1.3185	0.5437
1.2146	16.0	320	1.3288	0.55
1.2146	17.0	340	1.3262	0.5563
1.2146	18.0	360	1.3142	0.55
1.2146	19.0	380	1.2982	0.5625
1.0644	20.0	400	1.2704	0.5625
1.0644	21.0	420	1.2862	0.55
1.0644	22.0	440	1.2941	0.55
1.0644	23.0	460	1.2876	0.5312
1.0644	24.0	480	1.3066	0.5625
1.0161	25.0	500	1.2734	0.55

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0