Modelpublic

paolopertino/mobilenet-finetuned-food101

sourceHugging Faceotherupdated 3y agoView on Hugging Face

0likes31downloads

Model Card

mobilenet-finetuned-food101

This model is a fine-tuned version of google/mobilenet_v2_1.0_224 on the food101 dataset. It achieves the following results on the evaluation set:

—Loss: 0.5518
—Accuracy: 0.821

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
—trainbatchsize: 128
—evalbatchsize: 128
—seed: 42
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 512
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_ratio: 0.1
—num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	6	1.9575	0.153
1.9536	2.0	12	1.8509	0.265
1.9536	3.0	18	1.7003	0.451
1.7915	4.0	24	1.5181	0.578
1.4994	5.0	30	1.3609	0.631
1.4994	6.0	36	1.2321	0.669
1.2203	7.0	42	1.0696	0.69
1.2203	8.0	48	0.9676	0.723
1.0215	9.0	54	0.8888	0.729
0.8462	10.0	60	0.8380	0.74
0.8462	11.0	66	0.7461	0.778
0.744	12.0	72	0.6724	0.792
0.744	13.0	78	0.7314	0.769
0.6496	14.0	84	0.6831	0.77
0.6143	15.0	90	0.5937	0.81
0.6143	16.0	96	0.6217	0.793
0.5468	17.0	102	0.5965	0.788
0.5468	18.0	108	0.5944	0.813
0.5428	19.0	114	0.5869	0.812
0.5193	20.0	120	0.5565	0.82
0.5193	21.0	126	0.6155	0.803
0.4902	22.0	132	0.5685	0.817
0.4902	23.0	138	0.6097	0.789
0.4869	24.0	144	0.6002	0.8
0.4745	25.0	150	0.5569	0.814
0.4745	26.0	156	0.5414	0.821
0.4653	27.0	162	0.5806	0.807
0.4653	28.0	168	0.5663	0.807
0.4543	29.0	174	0.5412	0.825
0.4575	30.0	180	0.5518	0.821

Framework versions

—Transformers 4.35.2
—Pytorch 2.1.0+cu118
—Datasets 2.15.0
—Tokenizers 0.15.0