segformer-b5-finetuned-HikingHD

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

• Loss: 0.1077
• Mean Iou: 0.6364
• Mean Accuracy: 0.9770
• Overall Accuracy: 0.9771
• Accuracy Unlabeled: nan
• Accuracy Traversable: 0.9758
• Accuracy Non-traversable: 0.9781
• Iou Unlabeled: 0.0
• Iou Traversable: 0.9493
• Iou Non-traversable: 0.9600
• Local Testing:
  • Average inference time: 0.9943107657962376

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: 2
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Unlabeled	Accuracy Traversable	Accuracy Non-traversable	Iou Unlabeled	Iou Traversable	Iou Non-traversable
0.1595	1.33	20	0.1789	0.9314	0.9657	0.9649	nan	0.9727	0.9588	nan	0.9240	0.9388
0.3593	2.67	40	0.1137	0.9429	0.9731	0.9709	nan	0.9911	0.9551	nan	0.9372	0.9485
0.1002	4.0	60	0.0979	0.9363	0.9661	0.9677	nan	0.9531	0.9791	nan	0.9282	0.9444
0.0348	5.33	80	0.0933	0.9442	0.9713	0.9717	nan	0.9675	0.9750	nan	0.9375	0.9509
0.0374	6.67	100	0.0884	0.9459	0.9714	0.9727	nan	0.9611	0.9817	nan	0.9391	0.9528
0.0447	8.0	120	0.0886	0.9491	0.9737	0.9743	nan	0.9684	0.9789	nan	0.9430	0.9553
0.0464	9.33	140	0.0790	0.9564	0.9783	0.9780	nan	0.9804	0.9762	nan	0.9514	0.9614
0.0421	10.67	160	0.0868	0.9540	0.9764	0.9768	nan	0.9733	0.9796	nan	0.9485	0.9596
0.0253	12.0	180	0.0887	0.9530	0.9756	0.9763	nan	0.9700	0.9812	nan	0.9472	0.9587
0.0364	13.33	200	0.0960	0.9494	0.9733	0.9745	nan	0.9638	0.9829	nan	0.9431	0.9558
0.0276	14.67	220	0.0980	0.9470	0.9717	0.9732	nan	0.9595	0.9840	nan	0.9402	0.9538
0.0279	16.0	240	0.0914	0.9534	0.9761	0.9765	nan	0.9725	0.9796	nan	0.9478	0.9590
0.026	17.33	260	0.0886	0.9557	0.9778	0.9777	nan	0.9792	0.9764	nan	0.9506	0.9609
0.0228	18.67	280	0.0888	0.9547	0.9775	0.9771	nan	0.9804	0.9745	nan	0.9495	0.9599
0.0259	20.0	300	0.0984	0.9505	0.9742	0.9750	nan	0.9679	0.9806	nan	0.9444	0.9565
0.0306	21.33	320	0.0890	0.9542	0.9763	0.9769	nan	0.9716	0.9811	nan	0.9487	0.9598
0.0305	22.67	340	0.0967	0.6352	0.9752	0.9762	nan	0.9669	0.9834	0.0	0.9468	0.9586
0.0219	24.0	360	0.0983	0.9538	0.9764	0.9767	nan	0.9735	0.9792	nan	0.9483	0.9593
0.023	25.33	380	0.0940	0.6368	0.9771	0.9774	nan	0.9743	0.9799	0.0	0.9499	0.9606
0.0217	26.67	400	0.0973	0.6360	0.9767	0.9768	nan	0.9758	0.9776	0.0	0.9486	0.9595
0.0267	28.0	420	0.1023	0.6360	0.9770	0.9768	nan	0.9792	0.9749	0.0	0.9487	0.9593
0.0202	29.33	440	0.0955	0.6376	0.9783	0.9780	nan	0.9802	0.9764	0.0	0.9514	0.9615
0.0225	30.67	460	0.1016	0.6360	0.9763	0.9768	nan	0.9727	0.9800	0.0	0.9484	0.9595
0.0288	32.0	480	0.1026	0.6354	0.9756	0.9763	nan	0.9697	0.9815	0.0	0.9473	0.9588
0.0209	33.33	500	0.0977	0.6370	0.9771	0.9776	nan	0.9735	0.9808	0.0	0.9502	0.9609
0.0202	34.67	520	0.1005	0.6367	0.9772	0.9773	nan	0.9762	0.9782	0.0	0.9497	0.9604
0.0194	36.0	540	0.1032	0.6365	0.9771	0.9772	nan	0.9766	0.9776	0.0	0.9495	0.9601
0.0165	37.33	560	0.1013	0.6373	0.9777	0.9778	nan	0.9769	0.9785	0.0	0.9508	0.9612
0.0226	38.67	580	0.1005	0.6367	0.9771	0.9773	nan	0.9752	0.9790	0.0	0.9497	0.9604
0.0206	40.0	600	0.1032	0.6369	0.9773	0.9775	nan	0.9757	0.9789	0.0	0.9501	0.9607
0.016	41.33	620	0.1007	0.6373	0.9776	0.9777	nan	0.9761	0.9790	0.0	0.9506	0.9611
0.012	42.67	640	0.1048	0.6364	0.9768	0.9771	nan	0.9748	0.9789	0.0	0.9493	0.9601
0.0261	44.0	660	0.1073	0.6364	0.9770	0.9771	nan	0.9760	0.9780	0.0	0.9493	0.9600
0.0125	45.33	680	0.1088	0.6357	0.9761	0.9765	nan	0.9727	0.9795	0.0	0.9479	0.9591
0.0259	46.67	700	0.1073	0.6365	0.9770	0.9772	nan	0.9760	0.9780	0.0	0.9494	0.9601
0.0173	48.0	720	0.1052	0.6366	0.9770	0.9772	nan	0.9755	0.9785	0.0	0.9495	0.9602
0.0162	49.33	740	0.1077	0.6364	0.9770	0.9771	nan	0.9758	0.9781	0.0	0.9493	0.9600

Framework versions

• Transformers 4.35.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 2.14.5
• Tokenizers 0.14.0