SegFormer_mit-b5_Final-Set4-Grayscale_Not-Augmented_4_lr0.0001

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: 0.0217
• Mean Iou: 0.9708
• Mean Accuracy: 0.9835
• Overall Accuracy: 0.9941
• Accuracy Background: 0.9965
• Accuracy Melt: 0.9584
• Accuracy Substrate: 0.9957
• Iou Background: 0.9940
• Iou Melt: 0.9288
• Iou Substrate: 0.9895

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
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 20
• mixed_precision_training: Native AMP

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
0.1107	0.8850	50	0.1152	0.8138	0.8439	0.9627	0.9781	0.5623	0.9914	0.9677	0.5412	0.9325
0.0564	1.7699	100	0.0520	0.9163	0.9432	0.9829	0.9967	0.8488	0.9841	0.9806	0.7963	0.9721
0.0296	2.6549	150	0.0270	0.9557	0.9821	0.9906	0.9916	0.9621	0.9928	0.9893	0.8939	0.9839
0.042	3.5398	200	0.0226	0.9619	0.9763	0.9922	0.9934	0.9384	0.9969	0.9917	0.9077	0.9862
0.0166	4.4248	250	0.0300	0.9616	0.9768	0.9904	0.9957	0.9446	0.9903	0.9872	0.9153	0.9823
0.0159	5.3097	300	0.0203	0.9658	0.9863	0.9931	0.9946	0.9701	0.9941	0.9923	0.9169	0.9883
0.0121	6.1947	350	0.0221	0.9645	0.9795	0.9928	0.9937	0.9480	0.9968	0.9923	0.9141	0.9872
0.0149	7.0796	400	0.0220	0.9648	0.9821	0.9930	0.9949	0.9565	0.9951	0.9930	0.9138	0.9874
0.0352	7.9646	450	0.0215	0.9658	0.9764	0.9933	0.9959	0.9361	0.9971	0.9935	0.9158	0.9880
0.0106	8.8496	500	0.0201	0.9696	0.9820	0.9939	0.9961	0.9535	0.9962	0.9938	0.9256	0.9892
0.0095	9.7345	550	0.0216	0.9674	0.9796	0.9936	0.9955	0.9463	0.9969	0.9936	0.9202	0.9886
0.009	10.6195	600	0.0209	0.9702	0.9821	0.9941	0.9966	0.9539	0.9960	0.9940	0.9273	0.9894
0.0106	11.5044	650	0.0211	0.9700	0.9830	0.9940	0.9964	0.9568	0.9958	0.9940	0.9266	0.9893
0.0099	12.3894	700	0.0217	0.9708	0.9835	0.9941	0.9965	0.9584	0.9957	0.9940	0.9288	0.9895

Framework versions

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