SegFormer_Clean_Set1_95images_mit-b5

This model is a fine-tuned version of nvidia/mit-b5 on the Hasano20/Clean_Set1_95images dataset. It achieves the following results on the evaluation set:

• Loss: 0.0390
• Mean Iou: 0.9468
• Mean Accuracy: 0.9733
• Overall Accuracy: 0.9860
• Accuracy Background: 0.9960
• Accuracy Melt: 0.9390
• Accuracy Substrate: 0.9850
• Iou Background: 0.9899
• Iou Melt: 0.8763
• Iou Substrate: 0.9743

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Background	Accuracy Melt	Accuracy Substrate	Iou Background	Iou Melt	Iou Substrate
0.3883	0.5882	10	0.7088	0.5294	0.6161	0.8428	0.8644	0.0	0.9840	0.8428	0.0	0.7456
0.6271	1.1765	20	0.4185	0.5763	0.6455	0.8828	0.9472	0.0011	0.9882	0.9297	0.0011	0.7980
0.1779	1.7647	30	0.2746	0.6105	0.6712	0.9000	0.9943	0.0499	0.9694	0.9534	0.0474	0.8307
0.228	2.3529	40	0.2865	0.6102	0.6723	0.8897	0.9635	0.0820	0.9716	0.9359	0.0692	0.8254
0.1099	2.9412	50	0.2432	0.6646	0.7305	0.9018	0.9879	0.2657	0.9380	0.9495	0.2073	0.8369
0.1448	3.5294	60	0.3321	0.5993	0.6606	0.8987	0.9744	0.0140	0.9934	0.9613	0.0139	0.8226
0.2412	4.1176	70	0.2053	0.6581	0.7115	0.9150	0.9906	0.1590	0.9850	0.9734	0.1485	0.8525
0.1585	4.7059	80	0.2824	0.7094	0.8614	0.8838	0.9775	0.8013	0.8055	0.9504	0.3927	0.7851
0.2025	5.2941	90	0.2405	0.7011	0.8139	0.8924	0.9982	0.6013	0.8423	0.9387	0.3501	0.8144
0.2516	5.8824	100	0.2134	0.7488	0.8852	0.9083	0.9937	0.8227	0.8391	0.9721	0.4533	0.8212
0.275	6.4706	110	0.2856	0.7243	0.8793	0.8910	0.9965	0.8484	0.7932	0.9543	0.4339	0.7848
0.0721	7.0588	120	0.1417	0.7758	0.8225	0.9428	0.9913	0.4956	0.9804	0.9789	0.4530	0.8955
0.1478	7.6471	130	0.1383	0.7811	0.8383	0.9412	0.9828	0.5588	0.9733	0.9715	0.4727	0.8992
0.0541	8.2353	140	0.1654	0.7353	0.7778	0.9368	0.9958	0.3461	0.9915	0.9805	0.3400	0.8854
0.1068	8.8235	150	0.1001	0.8481	0.8900	0.9607	0.9977	0.6982	0.9742	0.9813	0.6358	0.9272
0.0879	9.4118	160	0.1177	0.8272	0.8658	0.9568	0.9914	0.6186	0.9875	0.9798	0.5785	0.9232
0.0855	10.0	170	0.0929	0.8763	0.9444	0.9650	0.9910	0.8886	0.9537	0.9848	0.7113	0.9327
0.102	10.5882	180	0.0770	0.8935	0.9405	0.9715	0.9962	0.8565	0.9689	0.9851	0.7486	0.9468
0.1044	11.1765	190	0.1401	0.7868	0.8367	0.9441	0.9696	0.5446	0.9957	0.9672	0.4853	0.9080
0.0705	11.7647	200	0.0822	0.8836	0.9507	0.9674	0.9924	0.9057	0.9542	0.9853	0.7276	0.9380
0.0583	12.3529	210	0.0670	0.9102	0.9489	0.9757	0.9957	0.8760	0.9750	0.9841	0.7914	0.9550
0.0337	12.9412	220	0.0718	0.9048	0.9384	0.9751	0.9960	0.8389	0.9803	0.9858	0.7756	0.9530
0.0237	13.5294	230	0.0634	0.9106	0.9419	0.9769	0.9957	0.8467	0.9832	0.9878	0.7879	0.9562
0.2478	14.1176	240	0.0724	0.8949	0.9289	0.9726	0.9958	0.8103	0.9806	0.9855	0.7514	0.9478
0.0237	14.7059	250	0.0570	0.9230	0.9610	0.9790	0.9950	0.9124	0.9757	0.9861	0.8226	0.9604
0.0237	15.2941	260	0.0564	0.9251	0.9650	0.9798	0.9957	0.9248	0.9745	0.9887	0.8253	0.9612
0.0414	15.8824	270	0.0786	0.8738	0.8997	0.9693	0.9926	0.7107	0.9959	0.9893	0.6917	0.9405
0.0444	16.4706	280	0.0431	0.9383	0.9686	0.9840	0.9962	0.9269	0.9828	0.9908	0.8539	0.9702
0.0307	17.0588	290	0.0416	0.9438	0.9719	0.9855	0.9942	0.9350	0.9864	0.9900	0.8675	0.9741
0.0335	17.6471	300	0.0420	0.9402	0.9635	0.9846	0.9943	0.9062	0.9900	0.9900	0.8589	0.9716
0.0717	18.2353	310	0.0448	0.9375	0.9651	0.9837	0.9971	0.9144	0.9837	0.9891	0.8533	0.9702
0.0225	18.8235	320	0.0403	0.9405	0.9635	0.9847	0.9947	0.9058	0.9899	0.9904	0.8595	0.9716
0.0315	19.4118	330	0.0394	0.9444	0.9686	0.9855	0.9956	0.9230	0.9873	0.9901	0.8698	0.9732
0.0178	20.0	340	0.0390	0.9468	0.9733	0.9860	0.9960	0.9390	0.9850	0.9899	0.8763	0.9743

Framework versions

• Transformers 4.41.2
• Pytorch 2.0.1+cu117
• Datasets 2.19.2
• Tokenizers 0.19.1