SegFormer_mit-b5_Clean-Set3_RGB

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

• Loss: 0.0207
• Mean Iou: 0.9744
• Mean Accuracy: 0.9865
• Overall Accuracy: 0.9940
• Accuracy Background: 0.9965
• Accuracy Melt: 0.9672
• Accuracy Substrate: 0.9957
• Iou Background: 0.9938
• Iou Melt: 0.9389
• Iou Substrate: 0.9905

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: 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: 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
0.3016	0.9434	50	0.2259	0.6885	0.7339	0.9268	0.9683	0.2451	0.9882	0.9455	0.2365	0.8834
0.1267	1.8868	100	0.1062	0.8505	0.9168	0.9620	0.9849	0.7996	0.9660	0.9706	0.6411	0.9398
0.0982	2.8302	150	0.0765	0.8725	0.9003	0.9718	0.9905	0.7183	0.9920	0.9803	0.6829	0.9544
0.0626	3.7736	200	0.0596	0.9124	0.9496	0.9793	0.9921	0.8731	0.9836	0.9824	0.7879	0.9668
0.0601	4.7170	250	0.0776	0.8931	0.9394	0.9733	0.9814	0.8536	0.9834	0.9762	0.7466	0.9566
0.0662	5.6604	300	0.0548	0.9176	0.9660	0.9803	0.9919	0.9280	0.9781	0.9875	0.7993	0.9662
0.0297	6.6038	350	0.0353	0.9452	0.9791	0.9872	0.9918	0.9581	0.9875	0.9895	0.8670	0.9792
0.0197	7.5472	400	0.0422	0.9332	0.9520	0.9853	0.9949	0.8670	0.9940	0.9899	0.8343	0.9753
0.0274	8.4906	450	0.0281	0.9589	0.9783	0.9904	0.9944	0.9475	0.9932	0.9913	0.9012	0.9843
0.0197	9.4340	500	0.0280	0.9569	0.9792	0.9901	0.9965	0.9507	0.9904	0.9920	0.8950	0.9836
0.0185	10.3774	550	0.0230	0.9644	0.9819	0.9918	0.9961	0.9564	0.9931	0.9923	0.9142	0.9867
0.0131	11.3208	600	0.0248	0.9663	0.9788	0.9922	0.9951	0.9449	0.9964	0.9922	0.9192	0.9874
0.0123	12.2642	650	0.0229	0.9682	0.9784	0.9926	0.9957	0.9424	0.9972	0.9931	0.9236	0.9879
0.0094	13.2075	700	0.0220	0.9673	0.9811	0.9925	0.9962	0.9519	0.9951	0.9930	0.9209	0.9878
0.0092	14.1509	750	0.0198	0.9721	0.9845	0.9935	0.9962	0.9617	0.9956	0.9933	0.9334	0.9895
0.0119	15.0943	800	0.0210	0.9688	0.9828	0.9928	0.9971	0.9571	0.9943	0.9932	0.9250	0.9883
0.0092	16.0377	850	0.0220	0.9688	0.9819	0.9928	0.9959	0.9543	0.9957	0.9929	0.9249	0.9885
0.0092	16.9811	900	0.0186	0.9718	0.9859	0.9934	0.9965	0.9666	0.9947	0.9936	0.9324	0.9894
0.0069	17.9245	950	0.0201	0.9725	0.9831	0.9936	0.9963	0.9564	0.9967	0.9937	0.9341	0.9898
0.011	18.8679	1000	0.0190	0.9742	0.9851	0.9939	0.9962	0.9628	0.9964	0.9937	0.9388	0.9903
0.009	19.8113	1050	0.0219	0.9714	0.9855	0.9933	0.9972	0.9652	0.9940	0.9936	0.9314	0.9891
0.0086	20.7547	1100	0.0199	0.9737	0.9872	0.9938	0.9961	0.9702	0.9953	0.9937	0.9373	0.9901
0.0086	21.6981	1150	0.0206	0.9737	0.9850	0.9938	0.9957	0.9625	0.9967	0.9936	0.9372	0.9902
0.0052	22.6415	1200	0.0205	0.9737	0.9866	0.9939	0.9960	0.9682	0.9957	0.9936	0.9372	0.9903
0.0079	23.5849	1250	0.0205	0.9745	0.9861	0.9940	0.9962	0.9658	0.9962	0.9937	0.9393	0.9905
0.0057	24.5283	1300	0.0210	0.9746	0.9849	0.9940	0.9961	0.9618	0.9968	0.9938	0.9397	0.9904
0.007	25.4717	1350	0.0212	0.9735	0.9858	0.9938	0.9963	0.9652	0.9957	0.9936	0.9369	0.9901
0.0059	26.4151	1400	0.0207	0.9744	0.9865	0.9940	0.9965	0.9672	0.9957	0.9938	0.9389	0.9905

Framework versions

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