segformer-b4-finetuned-UBC-two

This model is a fine-tuned version of tferhan/segformer-b1-finetuned-UBC on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2500
• Mean Iou: 0.4701
• Mean Accuracy: 0.7752
• Overall Accuracy: 0.7757
• Accuracy Background: nan
• Accuracy Residential: 0.7621
• Accuracy Non-residential: 0.7884
• Iou Background: 0.0
• Iou Residential: 0.7025
• Iou Non-residential: 0.7079

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: 4
• eval_batch_size: 4
• 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
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Background	Accuracy Residential	Accuracy Non-residential	Iou Background	Iou Residential	Iou Non-residential
0.6815	1.0	140	0.6220	0.4724	0.8023	0.8038	nan	0.7624	0.8421	0.0	0.6996	0.7176
0.3859	2.0	280	0.3193	0.4353	0.7151	0.7159	nan	0.6956	0.7347	0.0	0.6443	0.6615
0.293	3.0	420	0.2738	0.4382	0.7323	0.7340	nan	0.6863	0.7783	0.0	0.6376	0.6771
0.2697	4.0	560	0.2571	0.4279	0.7076	0.7065	nan	0.7356	0.6796	0.0	0.6563	0.6273
0.2517	5.0	700	0.2489	0.4453	0.7395	0.7413	nan	0.6923	0.7867	0.0	0.6551	0.6808
0.211	6.0	840	0.2387	0.4451	0.7356	0.7351	nan	0.7481	0.7231	0.0	0.6786	0.6566
0.2117	7.0	980	0.2348	0.4531	0.7430	0.7427	nan	0.7498	0.7361	0.0	0.6863	0.6731
0.1924	8.0	1120	0.2348	0.4649	0.7667	0.7678	nan	0.7389	0.7946	0.0	0.6884	0.7062
0.2213	9.0	1260	0.2401	0.4650	0.7665	0.7674	nan	0.7423	0.7906	0.0	0.6938	0.7011
0.2005	10.0	1400	0.2330	0.4534	0.7480	0.7484	nan	0.7377	0.7582	0.0	0.6796	0.6805
0.1723	11.0	1540	0.2453	0.4521	0.7484	0.7506	nan	0.6889	0.8079	0.0	0.6613	0.6951
0.1924	12.0	1680	0.2311	0.4576	0.7550	0.7545	nan	0.7695	0.7405	0.0	0.6963	0.6764
0.1843	13.0	1820	0.2418	0.4690	0.7711	0.7726	nan	0.7307	0.8115	0.0	0.6907	0.7165
0.1839	14.0	1960	0.2344	0.4631	0.7647	0.7658	nan	0.7362	0.7932	0.0	0.6845	0.7046
0.1698	15.0	2100	0.2378	0.4627	0.7604	0.7609	nan	0.7450	0.7758	0.0	0.6876	0.7006
0.1832	16.0	2240	0.2423	0.4594	0.7571	0.7585	nan	0.7182	0.7960	0.0	0.6796	0.6984
0.1501	17.0	2380	0.2432	0.4693	0.7717	0.7730	nan	0.7378	0.8056	0.0	0.6924	0.7156
0.1788	18.0	2520	0.2474	0.4785	0.7877	0.7886	nan	0.7647	0.8108	0.0	0.7134	0.7220
0.1524	19.0	2660	0.2513	0.4715	0.7809	0.7827	nan	0.7330	0.8287	0.0	0.6907	0.7237
0.1669	20.0	2800	0.2421	0.4685	0.7723	0.7726	nan	0.7659	0.7788	0.0	0.7036	0.7020
0.1539	21.0	2940	0.2449	0.4706	0.7736	0.7744	nan	0.7515	0.7956	0.0	0.6975	0.7142
0.1664	22.0	3080	0.2433	0.4685	0.7710	0.7716	nan	0.7549	0.7872	0.0	0.7001	0.7054
0.1717	23.0	3220	0.2463	0.4658	0.7679	0.7683	nan	0.7561	0.7797	0.0	0.6953	0.7020
0.1477	24.0	3360	0.2464	0.4671	0.7710	0.7717	nan	0.7513	0.7906	0.0	0.6960	0.7054
0.1491	25.0	3500	0.2457	0.4663	0.7675	0.7679	nan	0.7567	0.7782	0.0	0.6990	0.6998
0.1601	26.0	3640	0.2452	0.4699	0.7739	0.7746	nan	0.7554	0.7924	0.0	0.7015	0.7082
0.1436	27.0	3780	0.2538	0.4657	0.7683	0.7689	nan	0.7526	0.7841	0.0	0.6953	0.7018
0.1487	28.0	3920	0.2489	0.4668	0.7702	0.7707	nan	0.7567	0.7837	0.0	0.6986	0.7018
0.1514	29.0	4060	0.2456	0.4686	0.7730	0.7735	nan	0.7583	0.7877	0.0	0.6999	0.7060
0.1519	30.0	4200	0.2500	0.4701	0.7752	0.7757	nan	0.7621	0.7884	0.0	0.7025	0.7079

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.2