SegFormer-B0 Finetuned Segments Graffiti Open-source Image Segmentation Model - Accurately Identify and Segment Graffiti Areas in Images

Segformer B0 Finetuned Segments Graffiti

Developed by Adriatogi

An image segmentation model fine-tuned on graffiti segmentation datasets based on NVIDIA's SegFormer-B0 architecture, capable of accurately identifying and segmenting graffiti areas in images.

Image Segmentation

Transformers

Open Source License:Other #Graffiti Segmentation #High-Precision Segmentation #Urban Landscape Analysis

Downloads 14

Release Time : 3/19/2024

Model Overview

This model is specifically designed for graffiti detection and segmentation, capable of separating graffiti areas from the background in images, suitable for urban management, cultural heritage preservation, and other scenarios.

Model Features

High-Precision Graffiti Segmentation

Achieves a mean Intersection over Union (IoU) of 0.8048 and an average accuracy of 0.8943 on the evaluation dataset.

Lightweight Architecture

Based on the lightweight design of SegFormer-B0, suitable for deployment in resource-constrained environments.

Dual-Category Recognition

Accurately distinguishes between graffiti areas (0.9056 accuracy) and non-graffiti areas (0.8830 accuracy).

Model Capabilities

Image Segmentation

Graffiti Detection

Pixel-Level Classification

Use Cases

Urban Management

Illegal Graffiti Monitoring

Automatically detects illegal graffiti in urban areas to assist municipal cleaning efforts.

Accurately identifies graffiti areas with an IoU of 0.7870.

Cultural Heritage Preservation

Historical Building Graffiti Restoration

Identifies graffiti areas on historical buildings to assist restoration efforts.

Achieves an accuracy of 0.8830 in identifying non-graffiti areas.

🚀 segformer-b0-finetuned-segments-graffiti

This model is a fine - tuned version of [nvidia/mit - b0](https://huggingface.co/nvidia/mit - b0) on the Adriatogi/graffiti dataset. It can achieve excellent results in image segmentation tasks, providing high - precision recognition and classification for graffiti images.

🚀 Quick Start

This model is a fine - tuned version of [nvidia/mit - b0](https://huggingface.co/nvidia/mit - b0) on the Adriatogi/graffiti dataset. It achieves the following results on the evaluation set:

Loss: 0.3250
Mean Iou: 0.8048
Mean Accuracy: 0.8943
Overall Accuracy: 0.8929
Accuracy Not Graf: 0.8830
Accuracy Graf: 0.9056
Iou Not Graf: 0.8227
Iou Graf: 0.7870

📚 Documentation

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: 0.0001
train_batch_size: 1
eval_batch_size: 1
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.05
num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Not Graf	Accuracy Graf	Iou Not Graf	Iou Graf
0.5235	0.21	20	0.6135	0.6499	0.8016	0.7879	0.6926	0.9105	0.6476	0.6523
0.5744	0.42	40	0.4091	0.7237	0.8496	0.8398	0.7714	0.9279	0.7305	0.7169
0.3705	0.62	60	0.3959	0.7389	0.8592	0.8500	0.7864	0.9320	0.7469	0.7309
0.1897	0.83	80	0.3006	0.7748	0.8666	0.8774	0.9525	0.7807	0.8139	0.7357
0.1662	1.04	100	0.2900	0.7817	0.8723	0.8809	0.9407	0.8040	0.8164	0.7469
0.4537	1.25	120	0.2751	0.7956	0.8830	0.8886	0.9276	0.8384	0.8242	0.7669
0.1249	1.46	140	0.2719	0.7944	0.8841	0.8873	0.9094	0.8588	0.8196	0.7691
0.4985	1.67	160	0.3441	0.7463	0.8630	0.8550	0.7995	0.9264	0.7563	0.7363
0.4279	1.88	180	0.2911	0.7819	0.8764	0.8796	0.9016	0.8512	0.8082	0.7555
0.1776	2.08	200	0.2808	0.7928	0.8831	0.8864	0.9093	0.8569	0.8184	0.7673
0.209	2.29	220	0.2815	0.7857	0.8752	0.8832	0.9393	0.8111	0.8191	0.7522
0.152	2.5	240	0.2833	0.7921	0.8846	0.8854	0.8916	0.8775	0.8142	0.7700
0.5696	2.71	260	0.2698	0.8035	0.8921	0.8923	0.8941	0.8901	0.8238	0.7832
0.1003	2.92	280	0.3147	0.7739	0.8796	0.8729	0.8263	0.9329	0.7854	0.7624
0.1349	3.12	300	0.2961	0.7980	0.8906	0.8886	0.8747	0.9064	0.8154	0.7805
0.2552	3.33	320	0.2701	0.8001	0.8914	0.8900	0.8800	0.9028	0.8183	0.7820
0.1138	3.54	340	0.2808	0.7890	0.8854	0.8830	0.8664	0.9044	0.8065	0.7716
0.1602	3.75	360	0.2815	0.7956	0.8875	0.8875	0.8874	0.8875	0.8161	0.7751
0.0823	3.96	380	0.3195	0.7753	0.8799	0.8739	0.8325	0.9272	0.7879	0.7627
0.331	4.17	400	0.3339	0.7782	0.8821	0.8757	0.8312	0.9330	0.7901	0.7664
0.205	4.38	420	0.3083	0.7923	0.8885	0.8849	0.8595	0.9175	0.8077	0.7769
0.1659	4.58	440	0.3035	0.7887	0.8862	0.8826	0.8569	0.9156	0.8042	0.7731
0.1186	4.79	460	0.2856	0.8004	0.8839	0.8923	0.9500	0.8179	0.8323	0.7684
0.2964	5.0	480	0.3583	0.7592	0.8723	0.8633	0.8004	0.9442	0.7672	0.7512
0.0742	5.21	500	0.3269	0.7804	0.8820	0.8772	0.8444	0.9196	0.7947	0.7660
0.1355	5.42	520	0.3504	0.7784	0.8819	0.8759	0.8338	0.9301	0.7908	0.7661
0.0757	5.62	540	0.2771	0.8062	0.8927	0.8942	0.9050	0.8804	0.8280	0.7844
0.2015	5.83	560	0.3324	0.7851	0.8850	0.8802	0.8469	0.9232	0.7992	0.7711
0.1187	6.04	580	0.2853	0.8077	0.8943	0.8949	0.8995	0.8891	0.8282	0.7872
0.1243	6.25	600	0.3166	0.7968	0.8915	0.8875	0.8599	0.9232	0.8115	0.7820
0.0484	6.46	620	0.2876	0.8134	0.8968	0.8986	0.9110	0.8826	0.8349	0.7919
0.0772	6.67	640	0.2985	0.8085	0.8964	0.8951	0.8863	0.9064	0.8263	0.7907
0.2296	6.88	660	0.3134	0.8080	0.8951	0.8950	0.8940	0.8962	0.8274	0.7886
0.0544	7.08	680	0.3300	0.8014	0.8925	0.8907	0.8780	0.9070	0.8189	0.7839
0.0942	7.29	700	0.3133	0.8070	0.8936	0.8946	0.9013	0.8860	0.8280	0.7860
0.2432	7.5	720	0.3376	0.8014	0.8938	0.8905	0.8675	0.9201	0.8168	0.7860
0.0637	7.71	740	0.3021	0.8108	0.8968	0.8967	0.8965	0.8970	0.8301	0.7915
0.0946	7.92	760	0.3242	0.8048	0.8943	0.8929	0.8831	0.9054	0.8227	0.7870
0.1291	8.12	780	0.3315	0.8011	0.8934	0.8903	0.8689	0.9179	0.8169	0.7853
0.1077	8.33	800	0.3095	0.8117	0.8944	0.8979	0.9221	0.8667	0.8356	0.7877
0.177	8.54	820	0.3174	0.8117	0.8951	0.8977	0.9162	0.8740	0.8345	0.7888
0.057	8.75	840	0.3106	0.8111	0.8973	0.8968	0.8930	0.9016	0.8297	0.7925
0.2007	8.96	860	0.3645	0.7953	0.8909	0.8866	0.8571	0.9247	0.8097	0.7809
0.1281	9.17	880	0.3561	0.8008	0.8932	0.8902	0.8688	0.9176	0.8166	0.7850
0.0639	9.38	900	0.3120	0.8109	0.8969	0.8968	0.8962	0.8975	0.8301	0.7917
0.0766	9.58	920	0.3306	0.8057	0.8947	0.8934	0.8843	0.9051	0.8236	0.7877
0.1766	9.79	940	0.3321	0.8042	0.8941	0.8925	0.8813	0.9068	0.8219	0.7866
0.0842	10.0	960	0.3250	0.8048	0.8943	0.8929	0.8830	0.9056	0.8227	0.7870

Framework versions

Transformers 4.38.2
Pytorch 2.2.1+cu121
Datasets 2.18.0
Tokenizers 0.15.2

📄 License

other

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