emotion_face_image_classification_v2

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: 1.5157
• Accuracy: 0.4813

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: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_ratio: 0.1
• lr_scheduler_warmup_steps: 150
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.8	2	2.0924	0.15
No log	2.0	5	2.1024	0.0938
No log	2.8	7	2.0935	0.1375
No log	4.0	10	2.0893	0.15
No log	4.8	12	2.0900	0.15
No log	6.0	15	2.0987	0.0813
No log	6.8	17	2.0901	0.1
No log	8.0	20	2.0872	0.15
No log	8.8	22	2.0831	0.1375
No log	10.0	25	2.0750	0.1437
No log	10.8	27	2.0744	0.175
No log	12.0	30	2.0778	0.1437
No log	12.8	32	2.0729	0.1812
No log	14.0	35	2.0676	0.1625
No log	14.8	37	2.0694	0.1688
No log	16.0	40	2.0562	0.1625
No log	16.8	42	2.0498	0.1938
No log	18.0	45	2.0393	0.2188
No log	18.8	47	2.0458	0.2062
No log	20.0	50	2.0289	0.2125
No log	20.8	52	2.0226	0.2437
No log	22.0	55	1.9997	0.2625
No log	22.8	57	1.9855	0.3187
No log	24.0	60	1.9571	0.3187
No log	24.8	62	1.9473	0.3375
No log	26.0	65	1.9080	0.3187
No log	26.8	67	1.8894	0.35
No log	28.0	70	1.8407	0.375
No log	28.8	72	1.8083	0.3438
No log	30.0	75	1.7652	0.3563
No log	30.8	77	1.7281	0.3563
No log	32.0	80	1.6729	0.4062
No log	32.8	82	1.6527	0.3937
No log	34.0	85	1.6044	0.4562
No log	34.8	87	1.5899	0.4313
No log	36.0	90	1.5488	0.4313
No log	36.8	92	1.5340	0.45
No log	38.0	95	1.5227	0.4875
No log	38.8	97	1.4846	0.4875
No log	40.0	100	1.4579	0.4688

Framework versions

• Transformers 4.33.2
• Pytorch 2.0.1+cu118
• Datasets 2.14.5
• Tokenizers 0.13.3