Hasano20/SegFormer_mit-b5_Clean-Set3-Grayscale_Augmented_Soft

SegFormer_mit-b5_Clean-Set3-Grayscale_Soft-Augmented

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: 0.0143
• Mean Iou: 0.9787
• Mean Accuracy: 0.9898
• Overall Accuracy: 0.9944
• Accuracy Background: 0.9976
• Accuracy Melt: 0.9776
• Accuracy Substrate: 0.9943
• Iou Background: 0.9927
• Iou Melt: 0.9518
• Iou Substrate: 0.9915

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: 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
0.1578	0.5208	50	0.1042	0.8587	0.9293	0.9602	0.9866	0.8445	0.9568	0.9694	0.6723	0.9344
0.069	1.0417	100	0.0438	0.9400	0.9733	0.9845	0.9890	0.9435	0.9873	0.9817	0.8600	0.9783
0.1295	1.5625	150	0.0535	0.9282	0.9513	0.9807	0.9796	0.8764	0.9979	0.9774	0.8371	0.9700
0.0318	2.0833	200	0.0363	0.9495	0.9759	0.9867	0.9974	0.9462	0.9843	0.9851	0.8841	0.9792
0.041	2.6042	250	0.0290	0.9570	0.9763	0.9893	0.9918	0.9426	0.9946	0.9871	0.8987	0.9851
0.0281	3.125	300	0.0259	0.9628	0.9843	0.9898	0.9958	0.9691	0.9881	0.9859	0.9176	0.9849
0.0276	3.6458	350	0.0206	0.9669	0.9842	0.9917	0.9948	0.9643	0.9936	0.9907	0.9223	0.9878
0.0374	4.1667	400	0.0249	0.9628	0.9772	0.9907	0.9931	0.9421	0.9963	0.9903	0.9127	0.9854
0.056	4.6875	450	0.0197	0.9702	0.9847	0.9924	0.9943	0.9645	0.9952	0.9908	0.9309	0.9889
0.0092	5.2083	500	0.0171	0.9734	0.9889	0.9932	0.9961	0.9775	0.9932	0.9920	0.9384	0.9897
0.0217	5.7292	550	0.0183	0.9720	0.9880	0.9928	0.9968	0.9749	0.9923	0.9916	0.9355	0.9890
0.0185	6.25	600	0.0192	0.9720	0.9834	0.9929	0.9975	0.9580	0.9946	0.9920	0.9350	0.9891
0.0168	6.7708	650	0.0163	0.9748	0.9865	0.9935	0.9963	0.9680	0.9953	0.9920	0.9420	0.9903
0.014	7.2917	700	0.0160	0.9754	0.9850	0.9937	0.9960	0.9621	0.9968	0.9926	0.9431	0.9905
0.0133	7.8125	750	0.0253	0.9704	0.9842	0.9925	0.9963	0.9620	0.9942	0.9920	0.9308	0.9884
0.0103	8.3333	800	0.0168	0.9742	0.9852	0.9934	0.9977	0.9632	0.9946	0.9921	0.9405	0.9900
0.0142	8.8542	850	0.0139	0.9778	0.9868	0.9943	0.9974	0.9669	0.9959	0.9926	0.9494	0.9915
0.0112	9.375	900	0.0152	0.9761	0.9887	0.9938	0.9970	0.9748	0.9942	0.9927	0.9450	0.9905
0.0103	9.8958	950	0.0151	0.9769	0.9870	0.9940	0.9958	0.9687	0.9965	0.9928	0.9469	0.9909
0.0093	10.4167	1000	0.0133	0.9800	0.9898	0.9948	0.9967	0.9767	0.9960	0.9930	0.9546	0.9923
0.0108	10.9375	1050	0.0171	0.9763	0.9878	0.9931	0.9974	0.9733	0.9926	0.9898	0.9500	0.9890
0.0106	11.4583	1100	0.0141	0.9779	0.9880	0.9943	0.9972	0.9712	0.9956	0.9928	0.9494	0.9916
0.0118	11.9792	1150	0.0143	0.9787	0.9898	0.9944	0.9976	0.9776	0.9943	0.9927	0.9518	0.9915

Framework versions

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