SegFormer_mit-b5_Final-Set4-Grayscale_On-the-fly-Augmented_batch8_lr0.0002

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: 1.6332
• Mean Iou: 0.2042
• Mean Accuracy: 0.3547
• Overall Accuracy: 0.5005
• Accuracy Background: 0.8894
• Accuracy Melt: 0.0
• Accuracy Substrate: 0.1746
• Iou Background: 0.4549
• Iou Melt: 0.0
• Iou Substrate: 0.1576

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
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 25
• 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
0.0948	1.7699	50	3.2172	0.1562	0.3333	0.4687	1.0	0.0	0.0	0.4687	0.0	0.0
0.0735	3.5398	100	3.5520	0.1562	0.3333	0.4687	1.0	0.0	0.0	0.4687	0.0	0.0
0.064	5.3097	150	3.4135	0.1562	0.3333	0.4687	1.0	0.0	0.0	0.4687	0.0	0.0
0.062	7.0796	200	3.2473	0.1594	0.3297	0.4638	0.9688	0.0	0.0203	0.4585	0.0	0.0197
0.0672	8.8496	250	1.3897	0.1861	0.3555	0.5006	0.9884	0.0	0.0780	0.4812	0.0	0.0771
0.0423	10.6195	300	1.3204	0.1603	0.2938	0.4144	0.7495	0.0	0.1318	0.3750	0.0	0.1058
0.044	12.3894	350	1.2021	0.2482	0.3864	0.5467	0.8290	0.0	0.3303	0.4616	0.0	0.2829
0.0322	14.1593	400	1.5121	0.2118	0.3578	0.5052	0.8625	0.0	0.2108	0.4497	0.0	0.1858
0.0291	15.9292	450	1.6387	0.1855	0.3411	0.4808	0.9079	0.0	0.1155	0.4504	0.0	0.1059
0.0235	17.6991	500	1.6660	0.1874	0.3481	0.4906	0.9404	0.0	0.1040	0.4639	0.0	0.0982
0.0243	19.4690	550	1.5501	0.2051	0.3521	0.4970	0.8649	0.0	0.1913	0.4463	0.0	0.1690
0.0225	21.2389	600	1.7049	0.1982	0.3497	0.4934	0.8914	0.0	0.1578	0.4520	0.0	0.1427
0.0265	23.0088	650	1.6788	0.2008	0.3531	0.4982	0.8989	0.0	0.1606	0.4564	0.0	0.1461
0.0214	24.7788	700	1.6332	0.2042	0.3547	0.5005	0.8894	0.0	0.1746	0.4549	0.0	0.1576

Framework versions

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