segformer-b1-finetuned-cityscapes-1024-1024-straighter-only-test

This model is a fine-tuned version of nvidia/segformer-b1-finetuned-cityscapes-1024-1024 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0319
• Mean Iou: 0.9378
• Mean Accuracy: 0.9615
• Overall Accuracy: 0.9895
• Accuracy Default: 1e-06
• Accuracy Pipe: 0.8987
• Accuracy Floor: 0.9897
• Accuracy Background: 0.9959
• Iou Default: 1e-06
• Iou Pipe: 0.8434
• Iou Floor: 0.9813
• Iou Background: 0.9889

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: 3
• eval_batch_size: 3
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Default	Accuracy Pipe	Accuracy Floor	Accuracy Background	Iou Default	Iou Pipe	Iou Floor	Iou Background
0.3904	1.0	36	0.1465	0.8037	0.8484	0.9645	1e-06	0.5855	0.9696	0.9900	1e-06	0.5120	0.9355	0.9635
0.1244	2.0	72	0.0891	0.8640	0.9024	0.9766	1e-06	0.7371	0.9764	0.9938	1e-06	0.6565	0.9592	0.9762
0.0818	3.0	108	0.0669	0.8868	0.9178	0.9804	1e-06	0.7826	0.9745	0.9965	1e-06	0.7154	0.9657	0.9793
0.061	4.0	144	0.0525	0.9072	0.9407	0.9839	1e-06	0.8472	0.9801	0.9949	1e-06	0.7675	0.9711	0.9830
0.051	5.0	180	0.0470	0.9118	0.9444	0.9849	1e-06	0.8585	0.9790	0.9958	1e-06	0.7789	0.9722	0.9845
0.0461	6.0	216	0.0424	0.9191	0.9510	0.9861	1e-06	0.8736	0.9851	0.9944	1e-06	0.7959	0.9762	0.9851
0.0388	7.0	252	0.0401	0.9184	0.9443	0.9862	1e-06	0.8508	0.9862	0.9960	1e-06	0.7932	0.9769	0.9853
0.0348	8.0	288	0.0372	0.9244	0.9565	0.9870	1e-06	0.8894	0.9859	0.9943	1e-06	0.8104	0.9763	0.9865
0.0324	9.0	324	0.0362	0.9237	0.9486	0.9870	1e-06	0.8656	0.9833	0.9969	1e-06	0.8076	0.9773	0.9861
0.031	10.0	360	0.0349	0.9239	0.9520	0.9872	1e-06	0.8737	0.9870	0.9954	1e-06	0.8067	0.9788	0.9863
0.0287	11.0	396	0.0333	0.9285	0.9531	0.9877	1e-06	0.8720	0.9930	0.9944	1e-06	0.8209	0.9778	0.9868
0.0268	12.0	432	0.0332	0.9283	0.9522	0.9879	1e-06	0.8737	0.9865	0.9966	1e-06	0.8191	0.9787	0.9872
0.025	13.0	468	0.0311	0.9317	0.9622	0.9883	1e-06	0.9042	0.9877	0.9945	1e-06	0.8281	0.9794	0.9877
0.0247	14.0	504	0.0310	0.9308	0.9535	0.9884	1e-06	0.8742	0.9904	0.9959	1e-06	0.8247	0.9801	0.9876
0.0236	15.0	540	0.0307	0.9322	0.9538	0.9886	1e-06	0.8755	0.9897	0.9963	1e-06	0.8292	0.9793	0.9880
0.0223	16.0	576	0.0301	0.9346	0.9633	0.9888	1e-06	0.9083	0.9861	0.9955	1e-06	0.8360	0.9791	0.9886
0.0208	17.0	612	0.0308	0.9326	0.9578	0.9887	1e-06	0.8876	0.9907	0.9953	1e-06	0.8300	0.9797	0.9882
0.0198	18.0	648	0.0295	0.9339	0.9589	0.9888	1e-06	0.8897	0.9921	0.9949	1e-06	0.8335	0.9799	0.9882
0.0194	19.0	684	0.0311	0.9315	0.9524	0.9886	1e-06	0.8712	0.9894	0.9967	1e-06	0.8265	0.9802	0.9878
0.0188	20.0	720	0.0299	0.9332	0.9558	0.9888	1e-06	0.8807	0.9906	0.9959	1e-06	0.8318	0.9796	0.9882
0.0187	21.0	756	0.0298	0.9344	0.9567	0.9890	1e-06	0.8833	0.9905	0.9961	1e-06	0.8339	0.9810	0.9883
0.0179	22.0	792	0.0304	0.9334	0.9566	0.9889	1e-06	0.8834	0.9904	0.9959	1e-06	0.8317	0.9804	0.9882
0.0174	23.0	828	0.0301	0.9350	0.9603	0.9890	1e-06	0.8960	0.9895	0.9955	1e-06	0.8364	0.9803	0.9884
0.017	24.0	864	0.0294	0.9352	0.9589	0.9890	1e-06	0.8925	0.9877	0.9963	1e-06	0.8371	0.9802	0.9883
0.0172	25.0	900	0.0322	0.9334	0.9555	0.9888	1e-06	0.8796	0.9908	0.9960	1e-06	0.8320	0.9799	0.9882
0.0165	26.0	936	0.0312	0.9331	0.9556	0.9888	1e-06	0.8813	0.9891	0.9964	1e-06	0.8318	0.9792	0.9884
0.0162	27.0	972	0.0296	0.9350	0.9589	0.9891	1e-06	0.8911	0.9899	0.9959	1e-06	0.8360	0.9806	0.9885
0.0155	28.0	1008	0.0314	0.9359	0.9578	0.9892	1e-06	0.8880	0.9890	0.9965	1e-06	0.8384	0.9808	0.9884
0.0154	29.0	1044	0.0291	0.9379	0.9637	0.9894	1e-06	0.9061	0.9898	0.9952	1e-06	0.8438	0.9812	0.9887
0.0151	30.0	1080	0.0289	0.9372	0.9620	0.9893	1e-06	0.8994	0.9912	0.9952	1e-06	0.8419	0.9810	0.9887
0.0152	31.0	1116	0.0310	0.9365	0.9573	0.9893	1e-06	0.8865	0.9884	0.9969	1e-06	0.8397	0.9815	0.9884
0.0143	32.0	1152	0.0307	0.9376	0.9614	0.9894	1e-06	0.8983	0.9904	0.9956	1e-06	0.8433	0.9809	0.9887
0.0138	33.0	1188	0.0295	0.9385	0.9623	0.9896	1e-06	0.9004	0.9910	0.9955	1e-06	0.8451	0.9814	0.9889
0.0149	34.0	1224	0.0308	0.9380	0.9617	0.9894	1e-06	0.9007	0.9883	0.9961	1e-06	0.8444	0.9809	0.9886
0.0138	35.0	1260	0.0304	0.9376	0.9616	0.9894	1e-06	0.8993	0.9899	0.9958	1e-06	0.8431	0.9809	0.9888
0.0138	36.0	1296	0.0299	0.9379	0.9598	0.9895	1e-06	0.8932	0.9901	0.9962	1e-06	0.8433	0.9816	0.9887
0.0139	37.0	1332	0.0298	0.9378	0.9615	0.9895	1e-06	0.8983	0.9903	0.9958	1e-06	0.8435	0.9812	0.9889
0.0133	38.0	1368	0.0293	0.9393	0.9624	0.9897	1e-06	0.9008	0.9906	0.9958	1e-06	0.8467	0.9823	0.9889
0.0131	39.0	1404	0.0318	0.9368	0.9592	0.9893	1e-06	0.8922	0.9893	0.9963	1e-06	0.8406	0.9814	0.9884
0.0129	40.0	1440	0.0303	0.9382	0.9627	0.9895	1e-06	0.9034	0.9890	0.9958	1e-06	0.8447	0.9813	0.9887
0.0126	41.0	1476	0.0304	0.9392	0.9631	0.9896	1e-06	0.9037	0.9901	0.9956	1e-06	0.8471	0.9818	0.9887
0.0126	42.0	1512	0.0311	0.9378	0.9595	0.9895	1e-06	0.8929	0.9892	0.9965	1e-06	0.8432	0.9817	0.9887
0.0125	43.0	1548	0.0314	0.9383	0.9611	0.9895	1e-06	0.8974	0.9899	0.9960	1e-06	0.8453	0.9809	0.9888
0.0129	44.0	1584	0.0319	0.9374	0.9585	0.9895	1e-06	0.8886	0.9904	0.9964	1e-06	0.8420	0.9816	0.9887
0.0127	45.0	1620	0.0313	0.9380	0.9594	0.9895	1e-06	0.8920	0.9900	0.9964	1e-06	0.8436	0.9816	0.9887
0.0127	46.0	1656	0.0321	0.9379	0.9626	0.9895	1e-06	0.9029	0.9893	0.9957	1e-06	0.8444	0.9805	0.9890
0.0121	47.0	1692	0.0321	0.9377	0.9599	0.9895	1e-06	0.8930	0.9907	0.9960	1e-06	0.8430	0.9813	0.9888
0.0115	48.0	1728	0.0305	0.9390	0.9633	0.9897	1e-06	0.9043	0.9900	0.9957	1e-06	0.8463	0.9817	0.9890
0.0118	49.0	1764	0.0319	0.9378	0.9615	0.9895	1e-06	0.8987	0.9897	0.9959	1e-06	0.8434	0.9813	0.9889

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1
• Datasets 2.15.0
• Tokenizers 0.15.0