segformer-b1-finetuned-cityscapes-1024-1024-straighter-only

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: 0.0331
• Mean Iou: 0.9378
• Mean Accuracy: 0.9644
• Overall Accuracy: 0.9883
• Accuracy Default: 1e-06
• Accuracy Pipe: 0.9182
• Accuracy Floor: 0.9790
• Accuracy Background: 0.9961
• Iou Default: 1e-06
• Iou Pipe: 0.8600
• Iou Floor: 0.9637
• Iou Background: 0.9896

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: 3
• eval_batch_size: 3
• 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
0.5349	1.0	36	0.1593	0.8143	0.8613	0.9661	1e-06	0.6324	0.9614	0.9903	1e-06	0.5490	0.9277	0.9660
0.1472	2.0	72	0.0977	0.8792	0.9255	0.9782	1e-06	0.8153	0.9690	0.9922	1e-06	0.7119	0.9456	0.9800
0.0902	3.0	108	0.0708	0.9014	0.9285	0.9820	1e-06	0.8194	0.9690	0.9972	1e-06	0.7669	0.9558	0.9815
0.0662	4.0	144	0.0586	0.9146	0.9552	0.9842	1e-06	0.9036	0.9666	0.9954	1e-06	0.8015	0.9567	0.9856
0.0543	5.0	180	0.0490	0.9225	0.9514	0.9856	1e-06	0.8844	0.9734	0.9964	1e-06	0.8208	0.9606	0.9860
0.0486	6.0	216	0.0445	0.9252	0.9640	0.9862	1e-06	0.9244	0.9729	0.9947	1e-06	0.8265	0.9616	0.9875
0.042	7.0	252	0.0414	0.9279	0.9658	0.9867	1e-06	0.9315	0.9699	0.9959	1e-06	0.8332	0.9626	0.9880
0.0389	8.0	288	0.0381	0.9322	0.9695	0.9874	1e-06	0.9413	0.9716	0.9956	1e-06	0.8448	0.9632	0.9888
0.0359	9.0	324	0.0386	0.9319	0.9629	0.9871	1e-06	0.9215	0.9702	0.9970	1e-06	0.8451	0.9630	0.9877
0.034	10.0	360	0.0374	0.9313	0.9632	0.9873	1e-06	0.9202	0.9730	0.9965	1e-06	0.8422	0.9634	0.9883
0.0322	11.0	396	0.0383	0.9300	0.9570	0.9871	1e-06	0.8993	0.9746	0.9971	1e-06	0.8379	0.9642	0.9878
0.0306	12.0	432	0.0353	0.9340	0.9678	0.9876	1e-06	0.9358	0.9710	0.9965	1e-06	0.8494	0.9637	0.9888
0.0292	13.0	468	0.0337	0.9355	0.9734	0.9881	1e-06	0.9527	0.9719	0.9957	1e-06	0.8529	0.9637	0.9898
0.0286	14.0	504	0.0334	0.9355	0.9686	0.9881	1e-06	0.9352	0.9745	0.9960	1e-06	0.8530	0.9641	0.9895
0.0271	15.0	540	0.0325	0.9389	0.9682	0.9885	1e-06	0.9325	0.9758	0.9964	1e-06	0.8624	0.9648	0.9897
0.0266	16.0	576	0.0327	0.9373	0.9696	0.9883	1e-06	0.9378	0.9748	0.9961	1e-06	0.8576	0.9646	0.9897
0.0257	17.0	612	0.0350	0.9330	0.9673	0.9877	1e-06	0.9302	0.9766	0.9952	1e-06	0.8463	0.9636	0.9892
0.0246	18.0	648	0.0333	0.9354	0.9665	0.9881	1e-06	0.9269	0.9764	0.9961	1e-06	0.8522	0.9644	0.9896
0.0242	19.0	684	0.0326	0.9378	0.9681	0.9884	1e-06	0.9311	0.9772	0.9959	1e-06	0.8588	0.9648	0.9896
0.0231	20.0	720	0.0339	0.9366	0.9665	0.9883	1e-06	0.9256	0.9781	0.9958	1e-06	0.8557	0.9646	0.9896
0.0236	21.0	756	0.0333	0.9365	0.9702	0.9883	1e-06	0.9375	0.9779	0.9951	1e-06	0.8552	0.9644	0.9900
0.0227	22.0	792	0.0327	0.9375	0.9690	0.9885	1e-06	0.9339	0.9773	0.9958	1e-06	0.8577	0.9649	0.9900
0.0226	23.0	828	0.0331	0.9378	0.9644	0.9883	1e-06	0.9182	0.9790	0.9961	1e-06	0.8600	0.9637	0.9896

Framework versions

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