segformer-b5-finetuned-morphpadver1-hgo-coord-v8_mix_resample

This model is a fine-tuned version of nvidia/mit-b5 on the NICOPOI-9/morphpad_coord_hgo_512_4class_v2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2808
• Mean Iou: 0.8180
• Mean Accuracy: 0.8970
• Overall Accuracy: 0.8985
• Accuracy 0-0: 0.8794
• Accuracy 0-90: 0.9144
• Accuracy 90-0: 0.9102
• Accuracy 90-90: 0.8841
• Iou 0-0: 0.8354
• Iou 0-90: 0.8160
• Iou 90-0: 0.7873
• Iou 90-90: 0.8332

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: 6e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy 0-0	Accuracy 0-90	Accuracy 90-0	Accuracy 90-90	Iou 0-0	Iou 0-90	Iou 90-0	Iou 90-90
1.1426	1.3638	4000	1.1407	0.3066	0.4671	0.4780	0.3557	0.6489	0.5021	0.3617	0.2802	0.3465	0.3259	0.2737
0.4964	2.7276	8000	0.8038	0.4919	0.6535	0.6597	0.5503	0.7383	0.6895	0.6360	0.4907	0.4875	0.5034	0.4860
0.5578	4.0914	12000	0.5917	0.6309	0.7707	0.7728	0.7459	0.7610	0.8266	0.7496	0.6425	0.6311	0.6161	0.6340
0.2163	5.4552	16000	0.4977	0.6817	0.8092	0.8097	0.8572	0.8441	0.7854	0.7501	0.6897	0.6606	0.6859	0.6908
0.4687	6.8190	20000	0.3747	0.7571	0.8582	0.8597	0.8646	0.8490	0.9043	0.8149	0.7867	0.7648	0.7151	0.7618
0.1327	8.1827	24000	0.3118	0.7967	0.8845	0.8856	0.8731	0.9079	0.8814	0.8757	0.8107	0.7760	0.7862	0.8140
0.2156	9.5465	28000	0.2808	0.8180	0.8970	0.8985	0.8794	0.9144	0.9102	0.8841	0.8354	0.8160	0.7873	0.8332

Framework versions

• Transformers 4.48.3
• Pytorch 2.1.0
• Datasets 3.2.0
• Tokenizers 0.21.0