Modelpublic

argish/vit-base-patch16-224-in21k-facial-emotion-classification

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes6downloads

Model Card

vitferfinetuned

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

—Loss: 0.5470
—Accuracy: 0.8399
—F1 Macro: 0.8459
—F1 Micro: 0.8399
—Precision Macro: 0.8451
—Recall Macro: 0.8472
—F1 Angry: 0.7908
—Precision Angry: 0.7735
—Recall Angry: 0.8090
—F1 Disgust: 0.9321
—Precision Disgust: 0.9364
—Recall Disgust: 0.9279
—F1 Fear: 0.7547
—Precision Fear: 0.7677
—Recall Fear: 0.7422
—F1 Happy: 0.9386
—Precision Happy: 0.9556
—Recall Happy: 0.9222
—F1 Neutral: 0.8262
—Precision Neutral: 0.8105
—Recall Neutral: 0.8427
—F1 Sad: 0.7626
—Precision Sad: 0.7692
—Recall Sad: 0.7562
—F1 Surprise: 0.9164
—Precision Surprise: 0.9030
—Recall Surprise: 0.9302

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
—trainbatchsize: 32
—evalbatchsize: 32
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 64
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_ratio: 0.1
—num_epochs: 5
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 Macro	F1 Micro	Precision Macro	Recall Macro	F1 Angry	Precision Angry	Recall Angry	F1 Disgust	Precision Disgust	Recall Disgust	F1 Fear	Precision Fear	Recall Fear	F1 Happy	Precision Happy	Recall Happy	F1 Neutral	Precision Neutral	Recall Neutral	F1 Sad	Precision Sad	Recall Sad	F1 Surprise	Precision Surprise	Recall Surprise
1.0353	1.0	899	0.9539	0.6659	0.6128	0.6659	0.6136	0.6759	0.5457	0.6230	0.4854	0.4869	0.3312	0.9189	0.3857	0.6196	0.2800	0.8874	0.8872	0.8877	0.6550	0.5993	0.7220	0.5604	0.5187	0.6093	0.7683	0.7166	0.8281
0.7413	2.0	1798	0.7735	0.7400	0.7367	0.7400	0.7297	0.7520	0.6732	0.5915	0.7812	0.8264	0.7634	0.9009	0.5777	0.6519	0.5187	0.9088	0.9434	0.8767	0.7174	0.7219	0.7130	0.6216	0.6297	0.6137	0.8318	0.8059	0.8595
0.5094	3.0	2697	0.6381	0.7955	0.8039	0.7955	0.8033	0.8053	0.7633	0.7405	0.7875	0.9364	0.9450	0.9279	0.6834	0.6727	0.6945	0.9246	0.9425	0.9074	0.7696	0.7606	0.7788	0.6779	0.7035	0.6541	0.8723	0.8580	0.8871
0.4165	4.0	3596	0.5596	0.8319	0.8391	0.8319	0.8428	0.8366	0.7983	0.8326	0.7667	0.9401	0.9623	0.9189	0.7452	0.7219	0.7701	0.9327	0.9515	0.9145	0.8100	0.7809	0.8413	0.7446	0.7482	0.7410	0.9028	0.9023	0.9034
0.2772	5.0	4495	0.5151	0.8528	0.8575	0.8528	0.8582	0.8571	0.8179	0.7972	0.8396	0.9364	0.9450	0.9279	0.7839	0.8025	0.7662	0.9397	0.9486	0.9310	0.8335	0.8251	0.8421	0.7720	0.7696	0.7744	0.9190	0.9196	0.9184

Framework versions

—Transformers 4.47.0
—Pytorch 2.5.1+cu121
—Datasets 3.3.1
—Tokenizers 0.21.0