segformer-b0-finetuned-HikingHD

This model is a fine-tuned version of nvidia/mit-b0 on the twdent/HikingHD dataset. It achieves the following results on the evaluation set:

• Loss: 0.1189
• Mean Iou: 0.9224
• Mean Accuracy: 0.9627
• Overall Accuracy: 0.9622
• Accuracy Traversable: 0.9645
• Accuracy Non-traversable: 0.9608
• Iou Traversable: 0.9032
• Iou Non-traversable: 0.9415

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 Traversable	Accuracy Non-traversable	Iou Traversable	Iou Non-traversable
0.1555	1.33	20	0.3462	0.8817	0.9484	0.9401	0.9792	0.9176	0.8568	0.9067
0.1168	2.67	40	0.1551	0.8998	0.9529	0.9503	0.9628	0.9431	0.8764	0.9233
0.1054	4.0	60	0.1566	0.8910	0.9527	0.9452	0.9807	0.9247	0.8675	0.9146
0.0775	5.33	80	0.1892	0.8645	0.9415	0.9304	0.9830	0.9000	0.8378	0.8912
0.1111	6.67	100	0.1369	0.9015	0.9515	0.9514	0.9520	0.9511	0.8776	0.9255
0.0737	8.0	120	0.1358	0.9005	0.9503	0.9510	0.9476	0.9529	0.8761	0.9249
0.0908	9.33	140	0.1186	0.9097	0.9565	0.9556	0.9599	0.9532	0.8878	0.9316
0.0654	10.67	160	0.1177	0.9182	0.9624	0.9599	0.9719	0.9529	0.8986	0.9377
0.0871	12.0	180	0.1220	0.9105	0.9546	0.9563	0.9482	0.9609	0.8880	0.9330
0.0493	13.33	200	0.1237	0.9126	0.9559	0.9573	0.9504	0.9613	0.8907	0.9346
0.0643	14.67	220	0.1232	0.9107	0.9503	0.9567	0.9265	0.9741	0.8868	0.9345
0.0491	16.0	240	0.1199	0.9140	0.9573	0.9580	0.9545	0.9600	0.8926	0.9355
0.0556	17.33	260	0.1114	0.9199	0.9613	0.9609	0.9628	0.9598	0.9001	0.9396
0.0484	18.67	280	0.1137	0.9189	0.9628	0.9603	0.9720	0.9535	0.8995	0.9383
0.0607	20.0	300	0.1230	0.9163	0.9625	0.9588	0.9762	0.9488	0.8966	0.9359
0.044	21.33	320	0.1349	0.9077	0.9567	0.9545	0.9648	0.9485	0.8858	0.9297
0.0426	22.67	340	0.1313	0.9070	0.9563	0.9541	0.9646	0.9481	0.8850	0.9291
0.0269	24.0	360	0.1143	0.9226	0.9668	0.9620	0.9850	0.9487	0.9046	0.9406
0.0593	25.33	380	0.1038	0.9235	0.9616	0.9629	0.9570	0.9662	0.9041	0.9428
0.0321	26.67	400	0.1136	0.9179	0.9598	0.9599	0.9595	0.9602	0.8976	0.9383
0.0752	28.0	420	0.1196	0.9194	0.9627	0.9606	0.9705	0.9548	0.9000	0.9388
0.0812	29.33	440	0.1253	0.9216	0.9665	0.9615	0.9854	0.9477	0.9035	0.9398
0.0329	30.67	460	0.1023	0.9294	0.9671	0.9657	0.9725	0.9618	0.9120	0.9467
0.035	32.0	480	0.0969	0.9282	0.9658	0.9651	0.9686	0.9631	0.9104	0.9460
0.0332	33.33	500	0.1086	0.9231	0.9620	0.9626	0.9598	0.9643	0.9038	0.9424
0.0343	34.67	520	0.0962	0.9312	0.9689	0.9666	0.9774	0.9603	0.9145	0.9480
0.0337	36.0	540	0.1072	0.9251	0.9649	0.9635	0.9703	0.9595	0.9067	0.9434
0.0367	37.33	560	0.1033	0.9302	0.9692	0.9660	0.9809	0.9574	0.9135	0.9470
0.0327	38.67	580	0.1014	0.9312	0.9681	0.9666	0.9734	0.9627	0.9143	0.9482
0.0293	40.0	600	0.1202	0.9207	0.9622	0.9613	0.9656	0.9588	0.9012	0.9401
0.0272	41.33	620	0.1113	0.9246	0.9634	0.9633	0.9637	0.9631	0.9058	0.9433
0.0304	42.67	640	0.1070	0.9253	0.9643	0.9637	0.9668	0.9619	0.9069	0.9438
0.037	44.0	660	0.1120	0.9228	0.9629	0.9624	0.9648	0.9610	0.9037	0.9419
0.0323	45.33	680	0.1132	0.9213	0.9615	0.9617	0.9609	0.9621	0.9016	0.9409
0.0281	46.67	700	0.1203	0.9199	0.9616	0.9609	0.9643	0.9589	0.9002	0.9396
0.0339	48.0	720	0.1124	0.9253	0.9646	0.9637	0.9683	0.9610	0.9070	0.9437
0.0289	49.33	740	0.1189	0.9224	0.9627	0.9622	0.9645	0.9608	0.9032	0.9415

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.1+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1