SegFormer-B3 Open-Source Image Segmentation Model - Free Deployment for Accurate Identification of Object Types such as Coastal Water Bodies

Segformer B3 Ade 512 512 Finetuned Coasttrain

Developed by peldrak

An image segmentation model fine-tuned on coastline datasets based on nvidia/segformer-b3-finetuned-ade-512-512, specialized in recognizing land cover types such as water bodies, whitewater, and sediments.

Image Segmentation

Transformers

Open Source License:Other #Coastline Segmentation #High-precision Semantic Segmentation #Remote Sensing Image Processing

Downloads 18

Release Time : 2/22/2024

Model Overview

This model is a semantic segmentation model optimized for coastline scenarios, capable of accurately identifying 7 types of land cover including water bodies, whitewater, sediments, and vegetation, suitable for remote sensing image analysis and similar applications.

Model Features

Coastline Scenario Optimization

Specifically optimized for coastline land cover features, with excellent performance on categories like water bodies and whitewater.

Multi-category Recognition

Capable of simultaneously identifying 7 types of coastline land cover, including water bodies (86.77% IoU) and whitewater (67.95% IoU).

High-resolution Processing

Supports 512×512 resolution input, preserving more image details.

Model Capabilities

Image Semantic Segmentation

Multi-category Land Cover Recognition

Remote Sensing Image Analysis

Coastline Change Monitoring

Use Cases

Environmental Monitoring

Coastline Erosion Monitoring

Regular analysis to identify changes in sediment distribution.

Sediment recognition IoU reaches 76.49%.

Disaster Assessment

Flood Area Identification

Quickly extract flood-affected areas.

Water body recognition accuracy is 93.52%.

🚀 segformer-b3-ade-512-512-finetuned-coastTrain

This model is a fine - tuned version of [nvidia/segformer - b3 - finetuned - ade - 512 - 512](https://huggingface.co/nvidia/segformer - b3 - finetuned - ade - 512 - 512) on the peldrak/coastTrain_512 - 512 dataset. It can achieve high - quality image segmentation results on relevant evaluation sets.

📚 Documentation

This model achieves the following results on the evaluation set:

Loss: 0.7613
Mean Iou: 0.7092
Mean Accuracy: 0.8104
Overall Accuracy: 0.8790
Accuracy Water: 0.9352
Accuracy Whitewater: 0.8067
Accuracy Sediment: 0.8732
Accuracy Other Natural Terrain: 0.5054
Accuracy Vegetation: 0.8997
Accuracy Development: 0.8714
Accuracy Unknown: 0.7814
Iou Water: 0.8677
Iou Whitewater: 0.6795
Iou Sediment: 0.7649
Iou Other Natural Terrain: 0.4259
Iou Vegetation: 0.7883
Iou Development: 0.7211
Iou Unknown: 0.7170

🔧 Technical Details

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 6e - 05
train_batch_size: 4
eval_batch_size: 4
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Water	Accuracy Whitewater	Accuracy Sediment	Accuracy Other Natural Terrain	Accuracy Vegetation	Accuracy Development	Accuracy Unknown	Iou Water	Iou Whitewater	Iou Sediment	Iou Other Natural Terrain	Iou Vegetation	Iou Development	Iou Unknown
1.7642	0.05	20	1.6699	0.1741	0.2887	0.4511	0.3629	0.3020	0.0122	0.0013	0.8998	0.1317	0.3106	0.3310	0.0708	0.0112	0.0013	0.3953	0.1007	0.3084
1.6158	0.11	40	1.3903	0.1804	0.2783	0.5516	0.6957	0.0198	0.1032	0.0000	0.9605	0.1077	0.0616	0.5309	0.0184	0.0965	0.0000	0.4589	0.0973	0.0606
1.3168	0.16	60	1.1710	0.2583	0.3483	0.6425	0.8324	0.0359	0.0669	0.0	0.9578	0.1157	0.4296	0.6688	0.0344	0.0630	0.0	0.5089	0.1094	0.4233
1.1024	0.22	80	1.0398	0.3143	0.4032	0.6865	0.8815	0.1083	0.1413	0.0	0.9619	0.2673	0.4620	0.6970	0.1041	0.1261	0.0	0.5725	0.2452	0.4549
1.0384	0.27	100	0.9307	0.3388	0.4315	0.7113	0.8919	0.0379	0.3662	0.0	0.9582	0.2753	0.4913	0.7137	0.0374	0.2526	0.0	0.6316	0.2550	0.4813
0.9056	0.32	120	0.8649	0.3988	0.5060	0.7415	0.9191	0.1051	0.4270	0.0	0.8743	0.7201	0.4965	0.7159	0.1038	0.3178	0.0	0.6739	0.4951	0.4849
1.1867	0.38	140	0.8470	0.4027	0.5076	0.7418	0.8363	0.0329	0.6586	0.0	0.9529	0.5761	0.4960	0.7494	0.0326	0.4722	0.0	0.6188	0.4643	0.4815
1.2778	0.43	160	0.8108	0.4419	0.5491	0.7656	0.8973	0.1895	0.5864	0.0	0.9145	0.7679	0.4885	0.7758	0.1848	0.4732	0.0	0.6536	0.5269	0.4791
0.8217	0.49	180	0.7507	0.4544	0.5750	0.7728	0.8801	0.1928	0.7543	0.0	0.8893	0.7924	0.5161	0.7912	0.1858	0.5100	0.0	0.6678	0.5398	0.4859
0.9801	0.54	200	0.7149	0.4827	0.5995	0.7819	0.9016	0.3829	0.7254	0.0	0.8848	0.7904	0.5117	0.8007	0.3571	0.5615	0.0	0.6774	0.4859	0.4966
0.7374	0.59	220	0.6885	0.4950	0.6159	0.7894	0.9068	0.3910	0.8448	0.0	0.8656	0.7859	0.5169	0.7839	0.3448	0.5749	0.0	0.6942	0.5774	0.4895
1.0931	0.65	240	0.6884	0.4889	0.6134	0.7885	0.9106	0.3515	0.8590	0.0	0.8554	0.8118	0.5059	0.7804	0.3041	0.5561	0.0	0.7017	0.5858	0.4941
0.7106	0.7	260	0.8052	0.4413	0.5511	0.7563	0.9400	0.3677	0.3526	0.0	0.8507	0.8081	0.5382	0.7137	0.3061	0.2825	0.0	0.6967	0.5709	0.5193
0.7133	0.76	280	0.6507	0.5368	0.6542	0.8106	0.8931	0.6564	0.8631	0.0	0.9353	0.6824	0.5491	0.8105	0.5066	0.5893	0.0	0.7211	0.5953	0.5350
0.5858	0.81	300	0.6587	0.5212	0.6453	0.7979	0.9158	0.6788	0.6528	0.0	0.8872	0.8580	0.5241	0.8226	0.5180	0.5725	0.0	0.6814	0.5427	0.5113
1.9447	0.86	320	0.6674	0.5300	0.6268	0.8098	0.9182	0.4960	0.7161	0.0	0.9369	0.7516	0.5691	0.8130	0.4323	0.6061	0.0	0.6974	0.6098	0.5515
0.6724	0.92	340	0.6814	0.5191	0.6635	0.7901	0.8573	0.6785	0.8412	0.0	0.8680	0.8745	0.5251	0.7797	0.5249	0.5634	0.0	0.7007	0.5512	0.5139
0.6738	0.97	360	0.6131	0.5509	0.6663	0.8173	0.9235	0.6190	0.8401	0.0	0.8883	0.8535	0.5396	0.8125	0.5326	0.6451	0.0	0.7219	0.6215	0.5229
0.7131	1.03	380	0.6163	0.5582	0.6734	0.8172	0.8994	0.7309	0.8445	0.0	0.9282	0.7832	0.5272	0.8228	0.5820	0.6580	0.0	0.7059	0.6249	0.5137
0.8373	1.08	400	0.6077	0.5569	0.6737	0.8216	0.9115	0.7745	0.8120	0.0	0.9381	0.7219	0.5579	0.8242	0.5463	0.6524	0.0	0.7230	0.6089	0.5435
0.7344	1.14	420	0.6830	0.5195	0.6628	0.7866	0.9541	0.6517	0.8366	0.0	0.7106	0.8978	0.5890	0.7943	0.5367	0.5415	0.0	0.6433	0.5580	0.5626
0.4357	1.19	440	0.5908	0.5761	0.6867	0.8298	0.9183	0.7028	0.8313	0.0	0.9032	0.8332	0.6181	0.8259	0.5906	0.6593	0.0	0.7274	0.6289	0.6005
0.3423	1.24	460	0.5857	0.5864	0.7004	0.8332	0.9054	0.7249	0.8626	0.0	0.8887	0.8351	0.6862	0.8155	0.6248	0.6348	0.0	0.7415	0.6395	0.6484
0.5952	1.3	480	0.6290	0.5592	0.6560	0.8170	0.9176	0.7224	0.7925	0.0	0.9532	0.6852	0.5208	0.8244	0.6165	0.6594	0.0	0.7002	0.6048	0.5091
0.7312	1.35	500	0.6103	0.5637	0.6829	0.8213	0.9303	0.7779	0.8384	0.0	0.8961	0.8243	0.5137	0.8265	0.6301	0.6668	0.0	0.7254	0.5968	0.5000
0.4683	1.41	520	0.6372	0.5599	0.6809	0.8165	0.9056	0.7946	0.8165	0.0	0.9156	0.8230	0.5109	0.8273	0.6414	0.6765	0.0	0.7153	0.5598	0.4988
0.3688	1.46	540	0.6608	0.5537	0.6561	0.8129	0.9313	0.7464	0.6851	0.0005	0.9196	0.7613	0.5485	0.7956	0.6248	0.5633	0.0005	0.7172	0.6407	0.5340
0.3681	1.51	560	0.5841	0.5800	0.7074	0.8296	0.9114	0.8080	0.8613	0.0000	0.8767	0.8714	0.6232	0.8230	0.6089	0.6616	0.0000	0.7287	0.6480	0.5899
0.455	1.57	580	0.6379	0.5682	0.6749	0.8230	0.9313	0.7373	...	...	...	...	...	...	...	...	...	...	...	...

📄 License

The license of this model is other.

Property	Details
Model Type	Segformer - B3 fine - tuned on coastTrain dataset
Training Data	peldrak/coastTrain_512 - 512 dataset

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご