segformer-b1-finetuned-cityscapes-1024-1024-full-ds

This model is a fine-tuned version of nvidia/segformer-b1-finetuned-cityscapes-1024-1024 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0506
• Mean Iou: 0.9137
• Mean Accuracy: 0.9561
• Overall Accuracy: 0.9831
• Accuracy Default: 1e-06
• Accuracy Pipe: 0.9020
• Accuracy Floor: 0.9742
• Accuracy Background: 0.9920
• Iou Default: 1e-06
• Iou Pipe: 0.7996
• Iou Floor: 0.9590
• Iou Background: 0.9824

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: 0.0006
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Default	Accuracy Pipe	Accuracy Floor	Accuracy Background	Iou Default	Iou Pipe	Iou Floor	Iou Background
0.2488	1.0	39	0.1108	0.8539	0.9260	0.9669	1e-06	0.8345	0.9681	0.9754	1e-06	0.6794	0.9185	0.9639
0.0768	2.0	78	0.0659	0.8845	0.9254	0.9772	1e-06	0.8239	0.9573	0.9951	1e-06	0.7287	0.9506	0.9741
0.0663	3.0	117	0.0588	0.8918	0.9320	0.9793	1e-06	0.8343	0.9687	0.9931	1e-06	0.7439	0.9540	0.9776
0.0562	4.0	156	0.0534	0.9000	0.9592	0.9806	1e-06	0.9237	0.9627	0.9912	1e-06	0.7654	0.9539	0.9808
0.0509	5.0	195	0.0512	0.9063	0.9492	0.9817	1e-06	0.8876	0.9660	0.9940	1e-06	0.7813	0.9569	0.9806
0.0456	6.0	234	0.0498	0.9058	0.9550	0.9819	1e-06	0.9037	0.9692	0.9920	1e-06	0.7783	0.9574	0.9817
0.0425	7.0	273	0.0493	0.9045	0.9515	0.9817	1e-06	0.8918	0.9709	0.9918	1e-06	0.7748	0.9576	0.9810
0.0402	8.0	312	0.0503	0.9074	0.9456	0.9821	1e-06	0.8722	0.9706	0.9939	1e-06	0.7833	0.9581	0.9810
0.0382	9.0	351	0.0501	0.9108	0.9471	0.9825	1e-06	0.8766	0.9702	0.9943	1e-06	0.7930	0.9581	0.9812
0.0402	10.0	390	0.0474	0.9122	0.9520	0.9830	1e-06	0.8907	0.9720	0.9933	1e-06	0.7959	0.9583	0.9824
0.0367	11.0	429	0.0497	0.9089	0.9571	0.9824	1e-06	0.9088	0.9705	0.9919	1e-06	0.7863	0.9585	0.9820
0.0355	12.0	468	0.0445	0.9191	0.9618	0.9843	1e-06	0.9202	0.9719	0.9933	1e-06	0.8132	0.9597	0.9844
0.033	13.0	507	0.0494	0.9114	0.9543	0.9828	1e-06	0.8965	0.9746	0.9918	1e-06	0.7943	0.9571	0.9827
0.0319	14.0	546	0.0471	0.9163	0.9542	0.9837	1e-06	0.8953	0.9740	0.9934	1e-06	0.8068	0.9585	0.9835
0.0304	15.0	585	0.0476	0.9167	0.9527	0.9839	1e-06	0.8911	0.9726	0.9944	1e-06	0.8070	0.9598	0.9834
0.0304	16.0	624	0.0492	0.9151	0.9498	0.9835	1e-06	0.8812	0.9744	0.9939	1e-06	0.8036	0.9585	0.9832
0.0297	17.0	663	0.0504	0.9147	0.9549	0.9834	1e-06	0.9003	0.9705	0.9939	1e-06	0.8023	0.9587	0.9830
0.03	18.0	702	0.0504	0.9123	0.9584	0.9830	1e-06	0.9103	0.9732	0.9917	1e-06	0.7953	0.9588	0.9828
0.0294	19.0	741	0.0483	0.9162	0.9553	0.9839	1e-06	0.8980	0.9749	0.9931	1e-06	0.8054	0.9596	0.9838
0.0295	20.0	780	0.0506	0.9137	0.9561	0.9831	1e-06	0.9020	0.9742	0.9920	1e-06	0.7996	0.9590	0.9824

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1
• Datasets 2.15.0
• Tokenizers 0.15.0