segformer-b1-GFB

This model is a fine-tuned version of nvidia/mit-b1 on the segments/GFB dataset. It achieves the following results on the evaluation set:

• Loss: 0.7123
• Mean Iou: 0.4066
• Mean Accuracy: 0.6407
• Overall Accuracy: 0.7120
• Accuracy Unlabeled: 0.7269
• Accuracy Gbm: 0.7533
• Accuracy Podo: 0.6292
• Accuracy Endo: 0.4534
• Iou Unlabeled: 0.6879
• Iou Gbm: 0.3563
• Iou Podo: 0.3016
• Iou Endo: 0.2805

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: 0.0005
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 400
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Unlabeled	Accuracy Gbm	Accuracy Podo	Accuracy Endo	Iou Unlabeled	Iou Gbm	Iou Podo	Iou Endo
1.0437	2.3256	100	0.9647	0.1635	0.3944	0.3972	0.3432	0.9566	0.2760	0.0016	0.3400	0.1810	0.1312	0.0016
1.0477	4.6512	200	0.8946	0.2654	0.4802	0.5921	0.5917	0.9183	0.3820	0.0288	0.5746	0.2494	0.2106	0.0272
1.0521	6.9767	300	0.9437	0.1971	0.4167	0.4435	0.4006	0.9666	0.2893	0.0102	0.3960	0.1802	0.2019	0.0102
1.044	9.3023	400	0.8855	0.2978	0.5250	0.5906	0.5760	0.9407	0.4162	0.1673	0.5634	0.2525	0.2320	0.1432
1.043	11.6279	500	0.8286	0.3415	0.6245	0.6202	0.5835	0.8697	0.7344	0.3106	0.5767	0.3479	0.2393	0.2022
1.0244	13.9535	600	0.8240	0.3151	0.5919	0.5622	0.5104	0.9238	0.6422	0.2912	0.5034	0.2529	0.2886	0.2157
1.0181	16.2791	700	0.8666	0.2833	0.5509	0.5232	0.4694	0.9362	0.5532	0.2447	0.4624	0.2342	0.2497	0.1871
1.0278	18.6047	800	0.8137	0.3482	0.5795	0.6439	0.6303	0.9038	0.5999	0.1841	0.6138	0.2765	0.3384	0.1640
1.0576	20.9302	900	0.8081	0.4165	0.6371	0.7289	0.7441	0.7589	0.7191	0.3263	0.7099	0.4185	0.2868	0.2508
1.0154	23.2558	1000	0.8564	0.3388	0.5937	0.6087	0.5925	0.7811	0.6043	0.3967	0.5627	0.2989	0.2231	0.2706
0.9911	25.5814	1100	0.7912	0.3697	0.6326	0.6392	0.6166	0.8302	0.6749	0.4087	0.5943	0.2993	0.2939	0.2914
0.9804	27.9070	1200	0.7709	0.3985	0.6464	0.6974	0.6874	0.8488	0.7391	0.3104	0.6659	0.3707	0.3099	0.2476
1.0381	30.2326	1300	0.7870	0.3218	0.6275	0.5519	0.4872	0.8913	0.7242	0.4072	0.4783	0.2642	0.2766	0.2682
0.9724	32.5581	1400	0.8018	0.3666	0.6549	0.6235	0.5878	0.8067	0.7556	0.4696	0.5716	0.3192	0.2645	0.3110
0.9812	34.8837	1500	0.7901	0.3407	0.6203	0.5876	0.5474	0.8960	0.5768	0.4611	0.5349	0.2594	0.2900	0.2784
0.9904	37.2093	1600	0.8698	0.3007	0.5293	0.5806	0.5723	0.8089	0.4723	0.2638	0.5450	0.2474	0.2128	0.1977
0.9809	39.5349	1700	0.7359	0.3870	0.6742	0.6538	0.6237	0.8505	0.7126	0.5098	0.6068	0.3316	0.2965	0.3133
0.9835	41.8605	1800	0.6637	0.4472	0.7122	0.7321	0.7221	0.8559	0.7314	0.5394	0.7020	0.4129	0.3383	0.3357
0.9693	44.1860	1900	0.7658	0.3631	0.6283	0.6547	0.6421	0.8321	0.6226	0.4163	0.6148	0.3350	0.2732	0.2295
0.9553	46.5116	2000	0.6570	0.4282	0.7234	0.7041	0.6783	0.8463	0.7924	0.5764	0.6634	0.4159	0.3154	0.3180
0.9579	48.8372	2100	0.7036	0.4030	0.6615	0.6909	0.6793	0.8306	0.7145	0.4218	0.6555	0.3455	0.3298	0.2812
0.9291	51.1628	2200	0.7865	0.3438	0.6064	0.6146	0.5937	0.8297	0.5907	0.4113	0.5702	0.2705	0.2910	0.2434
0.9129	53.4884	2300	0.6790	0.4314	0.7052	0.7182	0.7075	0.8079	0.7652	0.5404	0.6840	0.4149	0.3219	0.3047
0.9365	55.8140	2400	0.7916	0.3621	0.5994	0.6570	0.6591	0.7521	0.6349	0.3514	0.6255	0.3046	0.2695	0.2487
0.9193	58.1395	2500	0.6435	0.4449	0.7146	0.7302	0.7229	0.8068	0.7568	0.5720	0.6981	0.4204	0.3320	0.3290
0.9009	60.4651	2600	0.6424	0.4394	0.7117	0.7213	0.7065	0.8426	0.7664	0.5314	0.6865	0.3993	0.3441	0.3274
0.8991	62.7907	2700	0.6142	0.4516	0.7266	0.7392	0.7304	0.8494	0.7231	0.6034	0.7099	0.4150	0.3630	0.3186
0.9053	65.1163	2800	0.6652	0.4470	0.6885	0.7437	0.7524	0.8031	0.6762	0.5223	0.7202	0.3968	0.3475	0.3235
0.8173	67.4419	2900	0.7179	0.3810	0.6336	0.6698	0.6637	0.7913	0.6622	0.4171	0.6357	0.3190	0.2975	0.2718
0.945	69.7674	3000	0.6487	0.4442	0.6973	0.7385	0.7416	0.7962	0.7164	0.5350	0.7112	0.4082	0.3428	0.3147
0.8838	72.0930	3100	0.7113	0.3930	0.6474	0.6827	0.6784	0.8044	0.6459	0.4609	0.6496	0.3371	0.2970	0.2883
0.8921	74.4186	3200	0.6615	0.4231	0.6799	0.7145	0.7118	0.8045	0.7055	0.4978	0.6840	0.3764	0.3258	0.3060
0.8662	76.7442	3300	0.6605	0.4303	0.6886	0.7238	0.7243	0.8056	0.6833	0.5414	0.6968	0.3823	0.3321	0.3101
0.8779	79.0698	3400	0.6636	0.4293	0.6729	0.7259	0.7311	0.7904	0.6939	0.4761	0.6998	0.3753	0.3354	0.3068
0.9394	81.3953	3500	0.6623	0.4412	0.6789	0.7404	0.7513	0.7854	0.6833	0.4954	0.7155	0.3982	0.3380	0.3133
0.8704	83.7209	3600	0.6811	0.4265	0.6654	0.7272	0.7386	0.7762	0.6573	0.4896	0.7030	0.3744	0.3255	0.3029
0.9146	86.0465	3700	0.7154	0.4051	0.6390	0.7087	0.7233	0.7467	0.6305	0.4557	0.6845	0.3505	0.2980	0.2875
0.8139	88.3721	3800	0.7432	0.3867	0.6199	0.6916	0.7069	0.7281	0.6125	0.4322	0.6659	0.3320	0.2822	0.2666
0.8596	90.6977	3900	0.7091	0.4057	0.6419	0.7085	0.7207	0.7538	0.6455	0.4477	0.6836	0.3548	0.3001	0.2842
0.8117	93.0233	4000	0.7186	0.3985	0.6374	0.7012	0.7130	0.7491	0.6307	0.4569	0.6756	0.3476	0.2919	0.2788
0.8227	95.3488	4100	0.7220	0.4034	0.6334	0.7106	0.7287	0.7434	0.6130	0.4485	0.6878	0.3494	0.2968	0.2798
0.7746	97.6744	4200	0.7147	0.4095	0.6381	0.7172	0.7356	0.7497	0.6206	0.4463	0.6946	0.3573	0.3042	0.2818
0.8396	100.0	4300	0.7123	0.4066	0.6407	0.7120	0.7269	0.7533	0.6292	0.4534	0.6879	0.3563	0.3016	0.2805

Framework versions

• Transformers 4.57.2
• Pytorch 2.8.0+cu126
• Datasets 4.4.1
• Tokenizers 0.22.1