dinov2-small-fer

This model is a fine-tuned version of facebook/dinov2-small-imagenet1k-1-layer on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8495
• Accuracy: 0.6964
• F1 angry: 0.6087
• F1 disgust: 0.5556
• F1 fear: 0.5091
• F1 happy: 0.8972
• F1 neutral: 0.6705
• F1 sad: 0.5893
• F1 surprise: 0.7923

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: 1e-05
• train_batch_size: 64
• eval_batch_size: 64
• 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: linear
• num_epochs: 12

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 angry	F1 disgust	F1 fear	F1 happy	F1 neutral	F1 sad	F1 surprise
No log	1.0	449	1.1298	0.5659	0.3647	0.0	0.3323	0.8466	0.5333	0.4221	0.6871
1.3025	2.0	898	0.9742	0.6322	0.5274	0.2029	0.4050	0.8734	0.6080	0.5073	0.7528
1.0413	3.0	1347	0.9270	0.6499	0.5517	0.4157	0.4662	0.8846	0.6287	0.4934	0.7694
0.954	4.0	1796	0.9256	0.6537	0.5383	0.3975	0.4553	0.8801	0.6342	0.5153	0.7714
0.8994	5.0	2245	0.8979	0.6655	0.5820	0.3896	0.4075	0.8881	0.6377	0.5610	0.7759
0.8501	6.0	2694	0.8805	0.6776	0.6003	0.4599	0.4657	0.8932	0.6599	0.5382	0.7816
0.7972	7.0	3143	0.9028	0.6723	0.5918	0.4094	0.3981	0.8890	0.6547	0.5487	0.7811
0.7676	8.0	3592	0.8530	0.6884	0.6152	0.5085	0.5093	0.8922	0.6667	0.5447	0.7933
0.7138	9.0	4041	0.8927	0.6761	0.5794	0.5771	0.4487	0.8852	0.6494	0.5670	0.7833
0.7138	10.0	4490	0.8556	0.6896	0.6114	0.5668	0.4927	0.8957	0.6613	0.5799	0.7859
0.6799	11.0	4939	0.8477	0.6973	0.6174	0.5672	0.5118	0.8974	0.6694	0.5843	0.7900
0.643	12.0	5388	0.8495	0.6964	0.6087	0.5556	0.5091	0.8972	0.6705	0.5893	0.7923

Framework versions

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