SegFormer_b2_

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: 1.9619
• Mean Iou: 0.6350
• Mean Accuracy: 0.7337
• Overall Accuracy: 0.9277
• Accuracy Road: 0.9852
• Accuracy Sidewalk: 0.8497
• Accuracy Building: 0.9358
• Accuracy Wall: 0.5532
• Accuracy Fence: 0.5198
• Accuracy Pole: 0.5455
• Accuracy Traffic light: 0.6870
• Accuracy Traffic sign: 0.7070
• Accuracy Vegetation: 0.9414
• Accuracy Terrain: 0.6285
• Accuracy Sky: 0.9488
• Accuracy Person: 0.7732
• Accuracy Rider: 0.5306
• Accuracy Car: 0.9299
• Accuracy Truck: 0.6569
• Accuracy Bus: 0.7685
• Accuracy Train: 0.6389
• Accuracy Motorcycle: 0.6015
• Accuracy Bicycle: 0.7396
• Iou Road: 0.9605
• Iou Sidewalk: 0.7409
• Iou Building: 0.8722
• Iou Wall: 0.4867
• Iou Fence: 0.4444
• Iou Pole: 0.3637
• Iou Traffic light: 0.4816
• Iou Traffic sign: 0.5829
• Iou Vegetation: 0.8786
• Iou Terrain: 0.5591
• Iou Sky: 0.9110
• Iou Person: 0.6011
• Iou Rider: 0.3690
• Iou Car: 0.8721
• Iou Truck: 0.5861
• Iou Bus: 0.7113
• Iou Train: 0.6222
• Iou Motorcycle: 0.4337
• Iou Bicycle: 0.5882

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	Accuracy Bicycle	Accuracy Building	Accuracy Bus	Accuracy Car	Accuracy Fence	Accuracy Motorcycle	Accuracy Person	Accuracy Pole	Accuracy Rider	Accuracy Road	Accuracy Sidewalk	Accuracy Sky	Accuracy Terrain	Accuracy Traffic light	Accuracy Traffic sign	Accuracy Train	Accuracy Truck	Accuracy Vegetation	Accuracy Wall	Iou Bicycle	Iou Building	Iou Bus	Iou Car	Iou Fence	Iou Motorcycle	Iou Person	Iou Pole	Iou Rider	Iou Road	Iou Sidewalk	Iou Sky	Iou Terrain	Iou Traffic light	Iou Traffic sign	Iou Train	Iou Truck	Iou Vegetation	Iou Wall	Validation Loss	Mean Accuracy	Mean Iou	Overall Accuracy
6.8123	2.1290	100	0.6984	0.9338	0.8122	0.9277	0.5514	0.5692	0.7411	0.5353	0.5676	0.9832	0.8382	0.9456	0.6402	0.6671	0.6933	0.6673	0.6179	0.9322	0.5422	0.5707	0.8673	0.7055	0.8721	0.4429	0.4264	0.5663	0.3490	0.3608	0.9568	0.7169	0.9046	0.5479	0.4801	0.5658	0.6310	0.5609	0.8722	0.4836	1.9557	0.7297	0.6253	0.9235
5.5587	4.2581	200	0.7256	0.9379	0.7677	0.9250	0.5431	0.5447	0.7266	0.5255	0.5780	0.9827	0.8446	0.9559	0.6846	0.6513	0.6983	0.7100	0.6409	0.9234	0.5933	0.5811	0.8659	0.6798	0.8731	0.4480	0.4214	0.5739	0.3477	0.3697	0.9569	0.7187	0.9050	0.5508	0.4799	0.5715	0.6571	0.5627	0.8720	0.5035	1.9616	0.7347	0.6284	0.9237
6.8729	6.3871	300	0.7543	0.9334	0.7977	0.9311	0.5201	0.4788	0.7494	0.5370	0.5464	0.9821	0.8502	0.9495	0.6429	0.6814	0.6879	0.6860	0.6341	0.9325	0.5513	0.5783	0.8654	0.7097	0.8738	0.4383	0.3938	0.5814	0.3495	0.3657	0.9587	0.7267	0.9026	0.5385	0.4817	0.5703	0.6481	0.5721	0.8733	0.4796	2.0128	0.7287	0.6267	0.9243
6.3252	8.5161	400	0.6642	0.9384	0.7729	0.9231	0.5929	0.4225	0.7956	0.5462	0.5785	0.9811	0.8564	0.9407	0.6866	0.6477	0.7079	0.7334	0.6215	0.9150	0.5680	0.5609	0.8657	0.6914	0.8719	0.4471	0.3463	0.5605	0.3529	0.3504	0.9575	0.7210	0.9026	0.5582	0.4800	0.5663	0.6632	0.5712	0.8702	0.4953	2.0218	0.7312	0.6228	0.9227
6.3303	10.6452	500	0.7177	0.9441	0.8494	0.9181	0.5622	0.5395	0.7004	0.5211	0.5410	0.9835	0.8381	0.9392	0.6200	0.6394	0.6761	0.6616	0.6112	0.9303	0.4396	0.5813	0.8632	0.6945	0.8714	0.4555	0.4227	0.5589	0.3463	0.3686	0.9574	0.7176	0.9037	0.5494	0.4400	0.5567	0.6219	0.5686	0.8730	0.4040	2.0294	0.7175	0.6187	0.9231
6.1549	12.7742	600	0.7508	0.9573	0.7949	0.9217	0.5561	0.6094	0.7175	0.4992	0.3451	0.9822	0.8461	0.9377	0.6346	0.6769	0.6496	0.6472	0.5465	0.8897	0.5458	0.5683	0.8503	0.6773	0.8694	0.4253	0.4378	0.5759	0.3383	0.2860	0.9581	0.7156	0.8943	0.5439	0.4632	0.5551	0.6143	0.5262	0.8543	0.4745	2.1125	0.7110	0.6120	0.9192
7.3083	14.9032	700	0.7100	0.9398	0.8242	0.9160	0.5903	0.6395	0.7402	0.5186	0.5756	0.9850	0.8476	0.9476	0.6635	0.6793	0.7236	0.6976	0.7515	0.9247	0.5481	0.5762	0.8665	0.7249	0.8740	0.4511	0.4341	0.5749	0.3518	0.3727	0.9589	0.7281	0.9005	0.5730	0.4763	0.5694	0.6395	0.6627	0.8741	0.4982	2.0793	0.7486	0.6372	0.9252
6.576	17.0215	800	0.7703	0.9409	0.7932	0.9190	0.5249	0.5203	0.7513	0.5143	0.5919	0.9821	0.8585	0.9530	0.6539	0.6359	0.6807	0.6271	0.6034	0.9322	0.4577	0.5755	0.8684	0.6832	0.8710	0.4431	0.3965	0.5864	0.3538	0.3662	0.9589	0.7242	0.9068	0.5658	0.4663	0.5700	0.5985	0.5316	0.8743	0.4129	2.0442	0.7216	0.6186	0.9247
6.1906	19.1505	900	0.6906	0.9467	0.8470	0.9224	0.5170	0.4961	0.7813	0.5332	0.5720	0.9847	0.8472	0.9447	0.6608	0.7185	0.7102	0.4138	0.5740	0.9189	0.4473	0.5695	0.8643	0.6888	0.8751	0.4150	0.3911	0.5883	0.3517	0.3777	0.9604	0.7371	0.9050	0.5748	0.4555	0.5705	0.4080	0.5304	0.8708	0.4035	2.0855	0.7119	0.6072	0.9242
7.41	21.2796	1000	0.7659	0.9404	0.7758	0.9210	0.5056	0.5361	0.7416	0.5777	0.6048	0.9830	0.8693	0.9475	0.6439	0.6821	0.7203	0.6634	0.6090	0.9171	0.4731	0.5747	0.8631	0.7228	0.8708	0.4197	0.4150	0.5772	0.3588	0.3654	0.9599	0.7358	0.9058	0.5547	0.4700	0.5752	0.6282	0.5567	0.8737	0.4317	2.0099	0.7304	0.6242	0.9239
5.5115	23.4086	1100	0.7564	0.9446	0.7604	0.9148	0.6497	0.5345	0.7515	0.5375	0.5970	0.9812	0.8744	0.9407	0.6428	0.6566	0.6718	0.7071	0.6244	0.9197	0.4471	0.5694	0.8659	0.7115	0.8685	0.4391	0.4190	0.5891	0.3549	0.3717	0.9597	0.7354	0.9072	0.5640	0.4857	0.5729	0.6665	0.5744	0.8741	0.4060	2.0197	0.7322	0.6282	0.9244
7.0305	25.5376	1200	0.7267	0.9504	0.7640	0.9260	0.5518	0.5643	0.7378	0.5290	0.5458	0.9839	0.8701	0.9395	0.6378	0.6620	0.6745	0.7368	0.6248	0.9167	0.5230	0.5713	0.8643	0.7156	0.8710	0.4571	0.4256	0.5882	0.3598	0.3668	0.9612	0.7406	0.9042	0.5554	0.4727	0.5726	0.6853	0.5750	0.8728	0.4675	1.9816	0.7297	0.6330	0.9258
5.2368	27.6667	1300	0.7217	0.9374	0.7948	0.9332	0.5699	0.5587	0.7443	0.5263	0.5490	0.9843	0.8599	0.9396	0.6362	0.6413	0.7024	0.7470	0.6401	0.9401	0.5039	0.5817	0.8697	0.7120	0.8732	0.4659	0.4127	0.5929	0.3582	0.3795	0.9610	0.7402	0.9039	0.5671	0.4764	0.5836	0.6913	0.5853	0.8782	0.4550	1.9868	0.7332	0.6362	0.9274
6.2172	29.7957	1400	0.7696	0.9371	0.7514	0.9329	0.5485	0.5655	0.7389	0.5560	0.5959	0.9858	0.8414	0.9414	0.6290	0.7111	0.6842	0.7097	0.5908	0.9342	0.5966	0.5896	0.8689	0.7058	0.8719	0.4575	0.4462	0.5838	0.3593	0.3561	0.9610	0.7383	0.9079	0.5570	0.4571	0.5676	0.6666	0.5487	0.8797	0.5196	2.0414	0.7379	0.6338	0.9268
5.8238	31.9247	1500	0.7155	0.9423	0.7631	0.9290	0.6080	0.5156	0.7468	0.5452	0.6095	0.9852	0.8504	0.9454	0.6533	0.6628	0.6916	0.6605	0.6024	0.9245	0.5715	0.5826	0.8678	0.7121	0.8697	0.4594	0.4105	0.5992	0.3578	0.3635	0.9609	0.7403	0.9070	0.5617	0.4771	0.5806	0.6362	0.5521	0.8765	0.5023	2.0376	0.7328	0.6325	0.9264
5.5947	34.0430	1600	0.7463	0.9465	0.8115	0.9198	0.5313	0.5075	0.7525	0.5320	0.5378	0.9834	0.8629	0.9352	0.6095	0.6516	0.6914	0.6838	0.6637	0.9242	0.5233	0.5755	0.8663	0.6922	0.8715	0.4515	0.4040	0.5862	0.3583	0.3677	0.9593	0.7355	0.9035	0.5453	0.4746	0.5805	0.6466	0.5881	0.8729	0.4686	1.9786	0.7271	0.6288	0.9253
6.8136	36.1720	1700	0.7539	0.9418	0.7692	0.9278	0.5493	0.5517	0.7664	0.5399	0.5394	0.9841	0.8605	0.9423	0.6269	0.6547	0.6952	0.6466	0.6442	0.9287	0.5673	0.5902	0.8702	0.7151	0.8725	0.4427	0.4200	0.5972	0.3619	0.3700	0.9602	0.7376	0.9088	0.5491	0.4791	0.5807	0.6186	0.5809	0.8764	0.4945	2.0119	0.7311	0.6329	0.9268
4.8895	39.0860	1800	1.9619	0.6350	0.7337	0.9277	0.9852	0.8497	0.9358	0.5532	0.5198	0.5455	0.6870	0.7070	0.9414	0.6285	0.9488	0.7732	0.5306	0.9299	0.6569	0.7685	0.6389	0.6015	0.7396	0.9605	0.7409	0.8722	0.4867	0.4444	0.3637	0.4816	0.5829	0.8786	0.5591	0.9110	0.6011	0.3690	0.8721	0.5861	0.7113	0.6222	0.4337	0.5882

Framework versions

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