segformer-b4-finetuned-morphpadver1-hgo-coord-v5

This model is a fine-tuned version of nvidia/mit-b4 on the NICOPOI-9/morphpad_coord_hgo_512_4class_v2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0305
• Mean Iou: 0.9957
• Mean Accuracy: 0.9978
• Overall Accuracy: 0.9978
• Accuracy 0-0: 0.9985
• Accuracy 0-90: 0.9967
• Accuracy 90-0: 0.9974
• Accuracy 90-90: 0.9988
• Iou 0-0: 0.9963
• Iou 0-90: 0.9949
• Iou 90-0: 0.9946
• Iou 90-90: 0.9970

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: 50

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
1.1205	2.6525	4000	1.0224	0.3348	0.5016	0.5008	0.3938	0.4555	0.5489	0.6084	0.3529	0.3168	0.3227	0.3469
0.7254	5.3050	8000	0.6015	0.5633	0.7192	0.7188	0.6706	0.7324	0.7551	0.7186	0.6037	0.5466	0.5314	0.5714
0.3698	7.9576	12000	0.3707	0.7204	0.8368	0.8369	0.8560	0.8606	0.8261	0.8044	0.7447	0.6982	0.6938	0.7450
0.2039	10.6101	16000	0.2198	0.8391	0.9123	0.9123	0.9102	0.9321	0.9138	0.8931	0.8560	0.8255	0.8231	0.8516
0.0791	13.2626	20000	0.1565	0.9056	0.9504	0.9504	0.9499	0.9421	0.9554	0.9542	0.9165	0.8932	0.8953	0.9172
0.0587	15.9151	24000	0.1037	0.9456	0.9721	0.9721	0.9732	0.9693	0.9708	0.9749	0.9512	0.9402	0.9402	0.9509
0.0739	18.5676	28000	0.0545	0.9677	0.9836	0.9836	0.9836	0.9762	0.9869	0.9876	0.9728	0.9635	0.9640	0.9704
0.0443	21.2202	32000	0.0565	0.9716	0.9856	0.9856	0.9887	0.9885	0.9817	0.9835	0.9767	0.9675	0.9687	0.9735
0.1436	23.8727	36000	0.0484	0.9763	0.9880	0.9880	0.9898	0.9858	0.9876	0.9889	0.9808	0.9734	0.9730	0.9782
0.0681	26.5252	40000	0.0467	0.9831	0.9915	0.9915	0.9915	0.9904	0.9912	0.9929	0.9861	0.9821	0.9790	0.9852
0.0138	29.1777	44000	0.0357	0.9868	0.9934	0.9934	0.9930	0.9927	0.9932	0.9946	0.9879	0.9856	0.9842	0.9897
0.0292	31.8302	48000	0.0295	0.9898	0.9949	0.9949	0.9957	0.9942	0.9949	0.9947	0.9899	0.9894	0.9879	0.9923
0.0081	34.4828	52000	0.0262	0.9915	0.9957	0.9957	0.9958	0.9951	0.9958	0.9962	0.9912	0.9910	0.9901	0.9937
0.0061	37.1353	56000	0.0388	0.9905	0.9952	0.9952	0.9949	0.9939	0.9952	0.9969	0.9909	0.9900	0.9874	0.9936
0.006	39.7878	60000	0.0415	0.9929	0.9964	0.9964	0.9949	0.9963	0.9964	0.9982	0.9926	0.9927	0.9907	0.9955
0.0056	42.4403	64000	0.0301	0.9943	0.9972	0.9972	0.9975	0.9971	0.9966	0.9974	0.9961	0.9927	0.9931	0.9954
0.005	45.0928	68000	0.0213	0.9957	0.9978	0.9978	0.9982	0.9979	0.9976	0.9978	0.9968	0.9948	0.9946	0.9967
0.0041	47.7454	72000	0.0305	0.9957	0.9978	0.9978	0.9985	0.9967	0.9974	0.9988	0.9963	0.9949	0.9946	0.9970

Framework versions

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