SegFormer_b2_2

This model is a fine-tuned version of nvidia/segformer-b2-finetuned-cityscapes-1024-1024 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.9606
• Mean Iou: 0.6207
• Mean Accuracy: 0.7217
• Overall Accuracy: 0.9249
• Accuracy Road: 0.9844
• Accuracy Sidewalk: 0.8451
• Accuracy Building: 0.9446
• Accuracy Wall: 0.5047
• Accuracy Fence: 0.5505
• Accuracy Pole: 0.5247
• Accuracy Traffic light: 0.6721
• Accuracy Traffic sign: 0.7141
• Accuracy Vegetation: 0.9260
• Accuracy Terrain: 0.6527
• Accuracy Sky: 0.9398
• Accuracy Person: 0.7681
• Accuracy Rider: 0.5605
• Accuracy Car: 0.9182
• Accuracy Truck: 0.5968
• Accuracy Bus: 0.7881
• Accuracy Train: 0.5123
• Accuracy Motorcycle: 0.5706
• Accuracy Bicycle: 0.7386
• Iou Road: 0.9579
• Iou Sidewalk: 0.7271
• Iou Building: 0.8669
• Iou Wall: 0.4481
• Iou Fence: 0.4465
• Iou Pole: 0.3530
• Iou Traffic light: 0.4760
• Iou Traffic sign: 0.5823
• Iou Vegetation: 0.8765
• Iou Terrain: 0.5576
• Iou Sky: 0.9035
• Iou Person: 0.5932
• Iou Rider: 0.3732
• Iou Car: 0.8710
• Iou Truck: 0.5520
• Iou Bus: 0.7234
• Iou Train: 0.4947
• Iou Motorcycle: 0.4111
• Iou Bicycle: 0.5787

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Road	Accuracy Sidewalk	Accuracy Building	Accuracy Wall	Accuracy Fence	Accuracy Pole	Accuracy Traffic light	Accuracy Traffic sign	Accuracy Vegetation	Accuracy Terrain	Accuracy Sky	Accuracy Person	Accuracy Rider	Accuracy Car	Accuracy Truck	Accuracy Bus	Accuracy Train	Accuracy Motorcycle	Accuracy Bicycle	Iou Road	Iou Sidewalk	Iou Building	Iou Wall	Iou Fence	Iou Pole	Iou Traffic light	Iou Traffic sign	Iou Vegetation	Iou Terrain	Iou Sky	Iou Person	Iou Rider	Iou Car	Iou Truck	Iou Bus	Iou Train	Iou Motorcycle	Iou Bicycle
16.6751	2.1290	100	2.9556	0.5273	0.5860	0.9128	0.9832	0.7636	0.9520	0.3754	0.3619	0.3673	0.3866	0.5925	0.9283	0.6073	0.8979	0.7549	0.2971	0.9528	0.3839	0.5714	0.1148	0.1847	0.6577	0.9480	0.6592	0.8484	0.3642	0.3387	0.2441	0.3544	0.5052	0.8692	0.5569	0.8726	0.5470	0.2646	0.8455	0.3804	0.5634	0.1148	0.1829	0.5595
14.9709	4.2581	200	2.3312	0.5652	0.6527	0.9067	0.9790	0.6836	0.9433	0.4577	0.4285	0.4349	0.5329	0.6837	0.9148	0.5706	0.9374	0.7045	0.5310	0.9232	0.4789	0.6725	0.3114	0.4178	0.7961	0.9282	0.5672	0.8545	0.4214	0.3899	0.2939	0.4030	0.5330	0.8612	0.5129	0.8959	0.5318	0.3426	0.8511	0.4733	0.6519	0.3102	0.3764	0.5404
12.5514	6.3871	300	2.2463	0.5818	0.6824	0.9099	0.9828	0.7372	0.9385	0.4104	0.5640	0.4339	0.6397	0.6520	0.9092	0.6454	0.9135	0.7698	0.5094	0.9151	0.5407	0.7326	0.5067	0.4544	0.7098	0.9406	0.6260	0.8504	0.3943	0.4213	0.2925	0.3770	0.5370	0.8563	0.5244	0.8847	0.5331	0.3304	0.8544	0.5204	0.7059	0.4756	0.3823	0.5481
11.8215	8.5161	400	2.1854	0.5677	0.6688	0.9123	0.9798	0.7777	0.9276	0.5374	0.5522	0.4754	0.6585	0.6626	0.9189	0.6628	0.9496	0.7286	0.5013	0.9211	0.4601	0.7198	0.2690	0.3037	0.7014	0.9419	0.6442	0.8586	0.4761	0.4304	0.3157	0.3894	0.5386	0.8601	0.4801	0.8998	0.5440	0.3374	0.8497	0.4551	0.6616	0.2649	0.2853	0.5536
12.7976	10.6452	500	2.1913	0.5681	0.6577	0.9132	0.9788	0.8357	0.9403	0.3648	0.4464	0.5153	0.5844	0.6711	0.9064	0.5650	0.9370	0.7434	0.5706	0.9340	0.4380	0.5866	0.5737	0.2666	0.6390	0.9501	0.6739	0.8517	0.3396	0.3917	0.3219	0.4339	0.5206	0.8594	0.5195	0.8937	0.5428	0.3339	0.8485	0.4335	0.5700	0.5384	0.2364	0.5346
13.2234	12.7742	600	2.1208	0.5839	0.6746	0.9159	0.9840	0.8034	0.9334	0.4014	0.5319	0.4315	0.5946	0.6512	0.9280	0.6488	0.9267	0.6349	0.5981	0.9371	0.4992	0.7449	0.4463	0.3991	0.7225	0.9498	0.6801	0.8532	0.3739	0.4087	0.3091	0.4380	0.5276	0.8633	0.5504	0.8880	0.5185	0.3530	0.8580	0.4889	0.7079	0.4403	0.3343	0.5509
9.7371	14.9032	700	2.1398	0.5802	0.6915	0.9116	0.9799	0.8241	0.9162	0.5241	0.5775	0.5446	0.6767	0.6958	0.9263	0.5810	0.9162	0.6330	0.5151	0.8971	0.5783	0.8181	0.3773	0.4174	0.7405	0.9472	0.6630	0.8494	0.4527	0.4216	0.3189	0.4272	0.5349	0.8626	0.4880	0.8874	0.5101	0.3422	0.8515	0.5549	0.6742	0.3667	0.3440	0.5263
8.4735	17.0215	800	2.1760	0.5756	0.6897	0.9129	0.9871	0.7981	0.9380	0.5897	0.6268	0.4616	0.6864	0.6777	0.8896	0.6343	0.9445	0.6620	0.6210	0.8976	0.5353	0.7562	0.3340	0.2657	0.7991	0.9503	0.6854	0.8490	0.4706	0.3476	0.3187	0.4378	0.5301	0.8558	0.5319	0.8957	0.5317	0.3490	0.8504	0.5175	0.6891	0.3327	0.2495	0.5435
7.8868	19.1505	900	2.0991	0.6117	0.7249	0.9189	0.9799	0.8504	0.9266	0.5342	0.6147	0.5391	0.6287	0.6909	0.9071	0.7484	0.9448	0.7911	0.5599	0.9310	0.6343	0.7929	0.5248	0.4726	0.7015	0.9576	0.7113	0.8583	0.4561	0.4635	0.3363	0.4626	0.5534	0.8615	0.4779	0.8956	0.5629	0.3528	0.8729	0.5908	0.7386	0.5084	0.4023	0.5594
9.4082	21.2796	1000	2.0767	0.5897	0.6879	0.9169	0.9831	0.8115	0.9418	0.4610	0.5540	0.5098	0.6356	0.6613	0.9094	0.6206	0.9356	0.7755	0.6345	0.9301	0.5099	0.7109	0.4752	0.4408	0.5701	0.9521	0.6884	0.8547	0.4137	0.4308	0.3356	0.4546	0.5351	0.8659	0.5416	0.8965	0.5601	0.3526	0.8540	0.4773	0.6614	0.4609	0.3696	0.4994
6.9498	23.4086	1100	2.1655	0.5859	0.6856	0.9165	0.9772	0.8440	0.9373	0.4009	0.6115	0.4860	0.5818	0.6818	0.9242	0.5435	0.9505	0.7132	0.5791	0.9152	0.4764	0.7693	0.4355	0.5341	0.6645	0.9516	0.6906	0.8536	0.3534	0.4365	0.3254	0.4377	0.5631	0.8669	0.4874	0.8886	0.5631	0.3553	0.8587	0.4529	0.6921	0.4264	0.3786	0.5496
7.1755	25.5376	1200	2.0846	0.5910	0.6962	0.9183	0.9765	0.8536	0.9404	0.5186	0.5747	0.5408	0.6658	0.6923	0.9122	0.5719	0.9314	0.7621	0.5426	0.9261	0.5853	0.7995	0.2802	0.4453	0.7091	0.9536	0.6891	0.8609	0.4342	0.4450	0.3391	0.4558	0.5683	0.8679	0.5108	0.8998	0.5627	0.3550	0.8639	0.5502	0.6730	0.2792	0.3643	0.5565
6.4426	27.6667	1300	2.0718	0.6111	0.7120	0.9207	0.9847	0.8011	0.9443	0.4555	0.5963	0.5093	0.6059	0.7045	0.9149	0.6177	0.9418	0.7521	0.5612	0.9359	0.5108	0.7845	0.6642	0.5255	0.7186	0.9539	0.6946	0.8617	0.4206	0.4598	0.3410	0.4472	0.5656	0.8704	0.5400	0.9032	0.5576	0.3591	0.8706	0.4810	0.7120	0.6166	0.4026	0.5532
7.9571	29.7957	1400	2.1007	0.6176	0.7312	0.9220	0.9852	0.8278	0.9265	0.4403	0.6432	0.5272	0.6923	0.7098	0.9338	0.6698	0.9471	0.7186	0.6121	0.9187	0.5939	0.7617	0.7162	0.5277	0.7412	0.9586	0.7228	0.8621	0.4078	0.4189	0.3411	0.4602	0.5732	0.8725	0.5544	0.9018	0.5617	0.3598	0.8695	0.5311	0.6980	0.6668	0.4134	0.5598
7.0214	31.9247	1500	2.0496	0.6164	0.7258	0.9226	0.9819	0.8393	0.9301	0.5485	0.5417	0.5299	0.6742	0.7202	0.9290	0.7253	0.9414	0.7524	0.5288	0.9318	0.5999	0.8061	0.5190	0.5778	0.7141	0.9578	0.7141	0.8628	0.4576	0.4463	0.3425	0.4592	0.5755	0.8727	0.5648	0.8974	0.5800	0.3549	0.8754	0.5577	0.7254	0.5025	0.4217	0.5440
7.6246	34.0430	1600	2.0286	0.6136	0.7167	0.9223	0.9861	0.8199	0.9443	0.5169	0.5326	0.5077	0.6671	0.7165	0.9199	0.6715	0.9394	0.6962	0.5626	0.9206	0.6453	0.7967	0.4386	0.5843	0.7517	0.9552	0.7119	0.8610	0.4590	0.4396	0.3443	0.4611	0.5784	0.8721	0.5778	0.8948	0.5691	0.3693	0.8707	0.5520	0.7157	0.4265	0.4399	0.5592
8.3015	36.1720	1700	2.0353	0.6179	0.7146	0.9237	0.9838	0.8350	0.9425	0.5130	0.5482	0.5338	0.6419	0.7140	0.9251	0.6892	0.9382	0.7216	0.5111	0.9258	0.6673	0.7389	0.6655	0.3349	0.7471	0.9571	0.7172	0.8667	0.4434	0.4582	0.3534	0.4672	0.5753	0.8749	0.5397	0.9001	0.5732	0.3670	0.8700	0.5908	0.6889	0.6197	0.3053	0.5727
6.6871	38.3011	1800	2.0154	0.6197	0.7235	0.9232	0.9853	0.8336	0.9387	0.5135	0.5580	0.5278	0.6452	0.7185	0.9232	0.6777	0.9465	0.7381	0.5634	0.9209	0.6325	0.7569	0.6470	0.5012	0.7192	0.9559	0.7157	0.8662	0.4477	0.4471	0.3452	0.4562	0.5791	0.8745	0.5668	0.9031	0.5781	0.3611	0.8668	0.5610	0.6953	0.6048	0.3805	0.5695
6.08	40.4301	1900	1.9764	0.6243	0.7265	0.9251	0.9837	0.8531	0.9402	0.5178	0.5974	0.5184	0.6617	0.6958	0.9265	0.6695	0.9388	0.7550	0.5007	0.9286	0.6250	0.7855	0.6083	0.5623	0.7346	0.9592	0.7311	0.8671	0.4530	0.4658	0.3482	0.4672	0.5793	0.8760	0.5586	0.8995	0.5776	0.3573	0.8744	0.5566	0.7267	0.5774	0.4128	0.5741
6.2931	42.5591	2000	1.9826	0.6251	0.7327	0.9242	0.9827	0.8623	0.9415	0.5115	0.5476	0.5138	0.6720	0.7138	0.9219	0.6876	0.9317	0.7541	0.5426	0.9236	0.5970	0.7784	0.6505	0.6230	0.7656	0.9588	0.7279	0.8642	0.4530	0.4526	0.3484	0.4763	0.5815	0.8742	0.5669	0.8976	0.5785	0.3623	0.8725	0.5541	0.7159	0.6131	0.4113	0.5689
7.358	44.6882	2100	1.9920	0.6200	0.7234	0.9252	0.9825	0.8570	0.9464	0.5359	0.5386	0.5277	0.6540	0.7065	0.9213	0.6690	0.9441	0.7394	0.5612	0.9279	0.6127	0.7689	0.5046	0.5876	0.7602	0.9600	0.7354	0.8655	0.4612	0.4359	0.3527	0.4766	0.5838	0.8746	0.5658	0.9032	0.5937	0.3755	0.8728	0.5446	0.7052	0.4878	0.4108	0.5756
6.1954	46.8172	2200	1.9624	0.6220	0.7258	0.9248	0.9841	0.8487	0.9439	0.5136	0.5489	0.5316	0.6693	0.7274	0.9223	0.6580	0.9433	0.7663	0.5655	0.9195	0.6042	0.7952	0.5396	0.5698	0.7396	0.9582	0.7289	0.8669	0.4539	0.4452	0.3528	0.4768	0.5826	0.8755	0.5567	0.9042	0.5920	0.3709	0.8711	0.5540	0.7234	0.5178	0.4115	0.5767
6.1774	48.9462	2300	1.9606	0.6207	0.7217	0.9249	0.9844	0.8451	0.9446	0.5047	0.5505	0.5247	0.6721	0.7141	0.9260	0.6527	0.9398	0.7681	0.5605	0.9182	0.5968	0.7881	0.5123	0.5706	0.7386	0.9579	0.7271	0.8669	0.4481	0.4465	0.3530	0.4760	0.5823	0.8765	0.5576	0.9035	0.5932	0.3732	0.8710	0.5520	0.7234	0.4947	0.4111	0.5787

Framework versions

• Transformers 4.47.1
• Pytorch 2.1.2+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0