segformer-b2-seed-67-v1

This model is a fine-tuned version of nvidia/mit-b3 on the unreal-hug/REAL_DATASET_SEG_331 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4746
• Mean Iou: 0.2841
• Mean Accuracy: 0.3507
• Overall Accuracy: 0.6084
• Accuracy Unlabeled: nan
• Accuracy Lv: 0.7915
• Accuracy Rv: 0.4646
• Accuracy Ra: 0.4834
• Accuracy La: 0.6858
• Accuracy Vs: 0.0
• Accuracy As: 0.0
• Accuracy Mk: 0.0
• Accuracy Tk: nan
• Accuracy Asd: 0.3160
• Accuracy Vsd: 0.2747
• Accuracy Ak: 0.4910
• Iou Unlabeled: 0.0
• Iou Lv: 0.7252
• Iou Rv: 0.4232
• Iou Ra: 0.4411
• Iou La: 0.5427
• Iou Vs: 0.0
• Iou As: 0.0
• Iou Mk: 0.0
• Iou Tk: nan
• Iou Asd: 0.2832
• Iou Vsd: 0.2342
• Iou Ak: 0.4759

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Unlabeled	Accuracy Lv	Accuracy Rv	Accuracy Ra	Accuracy La	Accuracy Vs	Accuracy As	Accuracy Mk	Accuracy Tk	Accuracy Asd	Accuracy Vsd	Accuracy Ak	Iou Unlabeled	Iou Lv	Iou Rv	Iou Ra	Iou La	Iou Vs	Iou As	Iou Mk	Iou Tk	Iou Asd	Iou Vsd	Iou Ak
1.2449	5.88	100	1.1508	0.1187	0.1954	0.4575	nan	0.8193	0.0533	0.1371	0.5424	0.0	0.0	0.0	nan	0.0171	0.0155	0.3697	0.0	0.5501	0.0518	0.1253	0.3509	0.0	0.0	0.0	0.0	0.0170	0.0148	0.3145
0.7118	11.76	200	0.7012	0.1534	0.2007	0.4466	nan	0.7352	0.1138	0.2300	0.5548	0.0	0.0	0.0	nan	0.0168	0.0284	0.3280	0.0	0.6079	0.1081	0.2084	0.4120	0.0	0.0	0.0	nan	0.0167	0.0276	0.3064
0.5567	17.65	300	0.5686	0.1896	0.2372	0.4810	nan	0.6994	0.2332	0.3522	0.5913	0.0	0.0	0.0	nan	0.0389	0.0765	0.3806	0.0	0.6382	0.2142	0.3023	0.4563	0.0	0.0	0.0	nan	0.0386	0.0714	0.3649
0.5054	23.53	400	0.5441	0.2473	0.3075	0.5803	nan	0.7991	0.4241	0.4885	0.5970	0.0	0.0	0.0	nan	0.1535	0.1388	0.4745	0.0	0.7215	0.3725	0.4107	0.4908	0.0	0.0	0.0	nan	0.1486	0.1228	0.4537
0.4344	29.41	500	0.5188	0.2706	0.3382	0.5967	nan	0.7810	0.4337	0.4668	0.7031	0.0	0.0	0.0	nan	0.2612	0.2644	0.4721	0.0	0.7121	0.3916	0.4164	0.5372	0.0	0.0	0.0	nan	0.2398	0.2236	0.4558
0.3796	35.29	600	0.5032	0.2669	0.3315	0.5911	nan	0.7953	0.4343	0.4050	0.6920	0.0	0.0	0.0	nan	0.2841	0.2321	0.4717	0.0	0.7196	0.3965	0.3778	0.5273	0.0	0.0	0.0	nan	0.2589	0.1996	0.4568
0.3888	41.18	700	0.4801	0.2798	0.3461	0.6037	nan	0.7862	0.4532	0.4667	0.6983	0.0	0.0	0.0	nan	0.3065	0.2590	0.4908	0.0	0.7192	0.4127	0.4292	0.5444	0.0	0.0	0.0	nan	0.2756	0.2216	0.4746
0.3467	47.06	800	0.4753	0.2822	0.3478	0.6061	nan	0.7919	0.4585	0.4857	0.6814	0.0	0.0	0.0	nan	0.3131	0.2640	0.4831	0.0	0.7259	0.4196	0.4424	0.5402	0.0	0.0	0.0	nan	0.2813	0.2262	0.4685
0.3757	52.94	900	0.4746	0.2841	0.3507	0.6084	nan	0.7915	0.4646	0.4834	0.6858	0.0	0.0	0.0	nan	0.3160	0.2747	0.4910	0.0	0.7252	0.4232	0.4411	0.5427	0.0	0.0	0.0	nan	0.2832	0.2342	0.4759
0.3616	58.82	1000	0.4788	0.2860	0.3537	0.6116	nan	0.7931	0.4687	0.4837	0.6922	0.0	0.0	0.0	nan	0.3193	0.2830	0.4970	0.0	0.7262	0.4259	0.4411	0.5449	0.0	0.0	0.0	nan	0.2856	0.2407	0.4817

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0