SegFormer_mit-b5_Clean-Set3-Grayscale

This model is a fine-tuned version of nvidia/mit-b5 on _Clean-Set3-Grayscale. It achieves the following results on the evaluation set:

• Train-Loss: 0.0053
• Loss: 0.0156
• Mean Iou: 0.9776
• Mean Accuracy: 0.9882
• Overall Accuracy: 0.9952
• Accuracy Background: 0.9974
• Accuracy Melt: 0.9708
• Accuracy Substrate: 0.9963
• Iou Background: 0.9942
• Iou Melt: 0.9458
• Iou Substrate: 0.9927

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 200
• num_epochs: 50

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
0.1206	1.8519	50	0.0898	0.8826	0.9277	0.9727	0.9809	0.8182	0.9840	0.9697	0.7209	0.9571
0.0687	3.7037	100	0.0445	0.9291	0.9568	0.9845	0.9920	0.8888	0.9895	0.9833	0.8286	0.9754
0.0457	5.5556	150	0.0413	0.9284	0.9428	0.9859	0.9938	0.8381	0.9966	0.9877	0.8204	0.9770
0.0281	7.4074	200	0.0240	0.9592	0.9706	0.9914	0.9971	0.9198	0.9949	0.9900	0.9011	0.9865
0.0234	9.2593	250	0.0179	0.9672	0.9810	0.9932	0.9960	0.9513	0.9957	0.9926	0.9195	0.9893
0.0147	11.1111	300	0.0180	0.9672	0.9785	0.9932	0.9955	0.9429	0.9972	0.9925	0.9197	0.9893
0.012	12.9630	350	0.0139	0.9748	0.9864	0.9946	0.9967	0.9664	0.9962	0.9936	0.9390	0.9918
0.0104	14.8148	400	0.0138	0.9756	0.9890	0.9947	0.9972	0.9748	0.9949	0.9935	0.9413	0.9919
0.0094	16.6667	450	0.0136	0.9767	0.9862	0.9950	0.9965	0.9646	0.9974	0.9940	0.9436	0.9924
0.0101	18.5185	500	0.0135	0.9767	0.9867	0.9950	0.9974	0.9663	0.9964	0.9940	0.9438	0.9924
0.0087	20.3704	550	0.0144	0.9764	0.9887	0.9949	0.9954	0.9736	0.9970	0.9935	0.9435	0.9923
0.0078	22.2222	600	0.0145	0.9760	0.9885	0.9949	0.9967	0.9727	0.9960	0.9938	0.9417	0.9924
0.0095	24.0741	650	0.0145	0.9753	0.9855	0.9948	0.9971	0.9626	0.9967	0.9939	0.9398	0.9921
0.0073	25.9259	700	0.0145	0.9761	0.9892	0.9949	0.9965	0.9752	0.9960	0.9938	0.9419	0.9925
0.009	27.7778	750	0.0143	0.9772	0.9891	0.9951	0.9958	0.9745	0.9970	0.9938	0.9451	0.9929
0.0049	29.6296	800	0.0143	0.9782	0.9883	0.9953	0.9966	0.9713	0.9971	0.9942	0.9474	0.9929
0.0075	31.4815	850	0.0153	0.9767	0.9886	0.9951	0.9967	0.9727	0.9963	0.9941	0.9434	0.9925
0.008	33.3333	900	0.0155	0.9772	0.9876	0.9952	0.9970	0.9690	0.9968	0.9943	0.9447	0.9927
0.0061	35.1852	950	0.0150	0.9777	0.9877	0.9953	0.9973	0.9691	0.9967	0.9943	0.9461	0.9928
0.0053	37.0370	1000	0.0156	0.9776	0.9882	0.9952	0.9974	0.9708	0.9963	0.9942	0.9458	0.9927

Framework versions

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