SegFormer_mit-b5_Clean-Set3-Grayscale_Augmented_Medium

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

• Train-Loss: 0.0088
• Val-Loss: 0.0134
• Mean Iou: 0.9793
• Mean Accuracy: 0.9903
• Overall Accuracy: 0.9947
• Accuracy Background: 0.9971
• Accuracy Melt: 0.9785
• Accuracy Substrate: 0.9952
• Iou Background: 0.9935
• Iou Melt: 0.9524
• Iou Substrate: 0.9920

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: 0.0002
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 25

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
0.1305	0.3968	50	0.1020	0.8694	0.9199	0.9644	0.9855	0.8016	0.9726	0.9651	0.6989	0.9443
0.0906	0.7937	100	0.0668	0.8972	0.9187	0.9757	0.9891	0.7703	0.9968	0.9818	0.7488	0.9609
0.0409	1.1905	150	0.0606	0.9231	0.9414	0.9814	0.9879	0.8379	0.9984	0.9840	0.8152	0.9702
0.0678	1.5873	200	0.0344	0.9524	0.9762	0.9879	0.9883	0.9463	0.9941	0.9848	0.8890	0.9834
0.0312	1.9841	250	0.0340	0.9489	0.9756	0.9874	0.9935	0.9442	0.9892	0.9869	0.8779	0.9818
0.0334	2.3810	300	0.0277	0.9576	0.9826	0.9895	0.9956	0.9637	0.9885	0.9908	0.8987	0.9833
0.0286	2.7778	350	0.0264	0.9581	0.9776	0.9898	0.9964	0.9452	0.9912	0.9896	0.9002	0.9846
0.0214	3.1746	400	0.0230	0.9661	0.9824	0.9915	0.9926	0.9587	0.9958	0.9903	0.9206	0.9875
0.0208	3.5714	450	0.0203	0.9692	0.9876	0.9922	0.9968	0.9751	0.9910	0.9916	0.9283	0.9878
0.0146	3.9683	500	0.0231	0.9667	0.9852	0.9915	0.9961	0.9680	0.9913	0.9904	0.9229	0.9870
0.0197	4.3651	550	0.0208	0.9662	0.9883	0.9916	0.9950	0.9790	0.9908	0.9914	0.9200	0.9873
0.0198	4.7619	600	0.0184	0.9722	0.9836	0.9930	0.9969	0.9587	0.9951	0.9916	0.9355	0.9896
0.019	5.1587	650	0.0211	0.9693	0.9889	0.9919	0.9970	0.9801	0.9896	0.9907	0.9298	0.9872
0.0115	5.5556	700	0.0193	0.9706	0.9833	0.9928	0.9963	0.9584	0.9953	0.9926	0.9304	0.9888
0.0135	5.9524	750	0.0166	0.9740	0.9867	0.9933	0.9965	0.9692	0.9945	0.9919	0.9401	0.9899
0.0127	6.3492	800	0.0182	0.9736	0.9866	0.9932	0.9969	0.9689	0.9939	0.9918	0.9395	0.9895
0.0129	6.7460	850	0.0194	0.9723	0.9853	0.9930	0.9958	0.9651	0.9951	0.9920	0.9354	0.9894
0.0124	7.1429	900	0.0145	0.9771	0.9900	0.9941	0.9972	0.9789	0.9940	0.9928	0.9472	0.9911
0.011	7.5397	950	0.0149	0.9774	0.9876	0.9941	0.9972	0.9704	0.9953	0.9923	0.9485	0.9914
0.0176	7.9365	1000	0.0212	0.9681	0.9890	0.9919	0.9972	0.9802	0.9895	0.9923	0.9251	0.9869
0.0205	8.3333	1050	0.0171	0.9724	0.9895	0.9930	0.9971	0.9797	0.9918	0.9924	0.9356	0.9893
0.0103	8.7302	1100	0.0141	0.9780	0.9891	0.9943	0.9968	0.9754	0.9953	0.9928	0.9497	0.9915
0.0093	9.1270	1150	0.0148	0.9769	0.9881	0.9941	0.9965	0.9723	0.9956	0.9930	0.9466	0.9911
0.0113	9.5238	1200	0.0136	0.9788	0.9881	0.9945	0.9977	0.9711	0.9955	0.9929	0.9517	0.9918
0.0132	9.9206	1250	0.0144	0.9783	0.9882	0.9944	0.9971	0.9720	0.9957	0.9930	0.9503	0.9915
0.0104	10.3175	1300	0.0135	0.9788	0.9882	0.9945	0.9976	0.9714	0.9957	0.9932	0.9515	0.9918
0.0153	10.7143	1350	0.0129	0.9796	0.9889	0.9947	0.9970	0.9734	0.9962	0.9932	0.9534	0.9922
0.0091	11.1111	1400	0.0142	0.9783	0.9900	0.9944	0.9968	0.9784	0.9950	0.9931	0.9500	0.9917
0.0098	11.5079	1450	0.0139	0.9789	0.9889	0.9946	0.9967	0.9740	0.9962	0.9933	0.9516	0.9920
0.0094	11.9048	1500	0.0136	0.9795	0.9887	0.9947	0.9977	0.9730	0.9956	0.9931	0.9533	0.9920
0.0088	12.3016	1550	0.0134	0.9793	0.9903	0.9947	0.9971	0.9785	0.9952	0.9935	0.9524	0.9920

Framework versions

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