Modelpublic

Shivagowri/vit-snacks

sourceHugging Faceapache-2.0updated 4y agoView on Hugging Face

0likes31downloads

Model Card

vit-snacks

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

—Loss: 0.2754
—Accuracy: 0.9393

Model description

upload any image of your fave yummy snack

Intended uses & limitations

there are only 20 different varieties of snacks

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

—learning_rate: 0.0002
—trainbatchsize: 16
—evalbatchsize: 8
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.8724	0.33	100	0.9118	0.8670
0.5628	0.66	200	0.6873	0.8471
0.4421	0.99	300	0.4995	0.8691
0.2837	1.32	400	0.4008	0.9026
0.1645	1.65	500	0.3702	0.9058
0.1604	1.98	600	0.3981	0.8921
0.0498	2.31	700	0.3185	0.9204
0.0406	2.64	800	0.3427	0.9141
0.1049	2.97	900	0.3444	0.9173
0.0272	3.3	1000	0.3168	0.9246
0.0186	3.63	1100	0.3142	0.9288
0.0203	3.96	1200	0.2931	0.9298
0.007	4.29	1300	0.2754	0.9393
0.0072	4.62	1400	0.2778	0.9403
0.0073	4.95	1500	0.2782	0.9393

Framework versions

—Transformers 4.20.1
—Pytorch 1.11.0+cu113
—Datasets 2.3.2
—Tokenizers 0.12.1