Segformer B4 Crack Segmentation Dataset

Developed by varcoder

A crack segmentation model based on the SegFormer architecture, fine-tuned on a crack segmentation dataset for detecting crack structures in images

Image Segmentation

Transformers

EnglishOpen Source License:Other #Crack Detection #High-Precision Segmentation #Building Structure Analysis

Downloads 200

Release Time : 7/12/2023

Model Overview

This model is an image segmentation model fine-tuned on the crack segmentation dataset based on nvidia/mit-b0, specifically designed to identify and segment crack regions in images

Model Features

Crack Detection

Optimized specifically for crack segmentation tasks, capable of accurately identifying crack structures in images

Lightweight Architecture

Based on the SegFormer-B4 architecture, it maintains performance while offering high computational efficiency

Fine-Tuning Optimization

Fine-tuned on a specific crack dataset, enhancing the ability to recognize crack features

Model Capabilities

Image Segmentation

Crack Detection

Structural Defect Identification

Use Cases

Infrastructure Inspection

Building Crack Detection

Used to detect cracks on building walls or floors

Achieved a crack accuracy of 0.6691 on the evaluation set

Road Crack Detection

Identifies crack defects on road surfaces

license: other tags:

generated_from_trainer model-index:
name: segformer-b4-crack-segmentation-dataset results: [] datasets:
varcoder/crack-segmentation-dataset language:
en library_name: transformers pipeline_tag: image-segmentation

segformer-b4-crack-segmentation-dataset

This model is a fine-tuned version of nvidia/mit-b0 on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.0594
Mean Iou: 0.3346
Mean Accuracy: 0.6691
Overall Accuracy: 0.6691
Accuracy Background: nan
Accuracy Crack: 0.6691
Iou Background: 0.0
Iou Crack: 0.6691

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: 6e-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Background	Accuracy Crack	Iou Crack
0.2287	0.02	100	0.2515	0.1734	0.3468	0.3468	nan	0.3468	0.3468
0.1792	0.04	200	0.1594	0.1671	0.3342	0.3342	nan	0.3342	0.3342
0.1177	0.06	300	0.1762	0.1044	0.2088	0.2088	nan	0.2088	0.2088
0.0821	0.08	400	0.1706	0.2065	0.4130	0.4130	nan	0.4130	0.4130
0.0666	0.1	500	0.1507	0.1931	0.3863	0.3863	nan	0.3863	0.3863
0.0675	0.12	600	0.1374	0.3114	0.6227	0.6227	nan	0.6227	0.6227
0.0267	0.15	700	0.1400	0.2171	0.4342	0.4342	nan	0.4342	0.4342
0.0192	0.17	800	0.1067	0.1594	0.3187	0.3187	nan	0.3187	0.3187
0.0711	0.19	900	0.1002	0.2915	0.5830	0.5830	nan	0.5830	0.5830
0.0761	0.21	1000	0.0785	0.3099	0.6199	0.6199	nan	0.6199	0.6199
0.0802	0.23	1100	0.0829	0.3086	0.6173	0.6173	nan	0.6173	0.6173
0.1058	0.25	1200	0.0895	0.2139	0.4278	0.4278	nan	0.4278	0.4278
0.0409	0.27	1300	0.0792	0.3237	0.6475	0.6475	nan	0.6475	0.6475
0.063	0.29	1400	0.0739	0.3084	0.6168	0.6168	nan	0.6168	0.6168
0.0669	0.31	1500	0.0747	0.3326	0.6653	0.6653	nan	0.6653	0.6653
0.1277	0.33	1600	0.0735	0.3149	0.6297	0.6297	nan	0.6297	0.6297
0.0388	0.35	1700	0.0708	0.2525	0.5050	0.5050	nan	0.5050	0.5050
0.0332	0.37	1800	0.0726	0.2908	0.5816	0.5816	nan	0.5816	0.5816
0.0435	0.4	1900	0.0673	0.2893	0.5786	0.5786	nan	0.5786	0.5786
0.1297	0.42	2000	0.0698	0.3438	0.6877	0.6877	nan	0.6877	0.6877
0.1202	0.44	2100	0.0745	0.2899	0.5798	0.5798	nan	0.5798	0.5798
0.0549	0.46	2200	0.0657	0.3522	0.7044	0.7044	nan	0.7044	0.7044
0.0223	0.48	2300	0.0808	0.2686	0.5372	0.5372	nan	0.5372	0.5372
0.0464	0.5	2400	0.0631	0.3221	0.6442	0.6442	nan	0.6442	0.6442
0.0364	0.52	2500	0.0778	0.3410	0.6820	0.6820	nan	0.6820	0.6820
0.047	0.54	2600	0.0689	0.3489	0.6978	0.6978	nan	0.6978	0.6978
0.0322	0.56	2700	0.0640	0.2863	0.5727	0.5727	nan	0.5727	0.5727
0.0453	0.58	2800	0.0574	0.3340	0.6681	0.6681	nan	0.6681	0.6681
0.0347	0.6	2900	0.0611	0.3289	0.6578	0.6578	nan	0.6578	0.6578
0.0916	0.62	3000	0.0609	0.3357	0.6714	0.6714	nan	0.6714	0.6714
0.0523	0.65	3100	0.0557	0.3318	0.6637	0.6637	nan	0.6637	0.6637
0.1246	0.67	3200	0.0558	0.3294	0.6588	0.6588	nan	0.6588	0.6588
0.0501	0.69	3300	0.0697	0.2955	0.5910	0.5910	nan	0.5910	0.5910
0.0312	0.71	3400	0.0604	0.3414	0.6827	0.6827	nan	0.6827	0.6827
0.0449	0.73	3500	0.0612	0.3305	0.6611	0.6611	nan	0.6611	0.6611
0.0111	0.75	3600	0.0617	0.2930	0.5860	0.5860	nan	0.5860	0.5860
0.0206	0.77	3700	0.0627	0.3663	0.7326	0.7326	nan	0.7326	0.7326
0.051	0.79	3800	0.0649	0.3159	0.6318	0.6318	nan	0.6318	0.6318
0.0243	0.81	3900	0.0600	0.3370	0.6740	0.6740	nan	0.6740	0.6740
0.0108	0.83	4000	0.0614	0.3595	0.7190	0.7190	nan	0.7190	0.7190
0.0951	0.85	4100	0.0564	0.3571	0.7142	0.7142	nan	0.7142	0.7142
0.0731	0.87	4200	0.0597	0.3497	0.6994	0.6994	nan	0.6994	0.6994
0.0307	0.9	4300	0.0636	0.3468	0.6937	0.6937	nan	0.6937	0.6937
0.1039	0.92	4400	0.0594	0.3397	0.6795	0.6795	nan	0.6795	0.6795
0.0083	0.94	4500	0.0606	0.3512	0.7024	0.7024	nan	0.7024	0.7024
0.0113	0.96	4600	0.0597	0.3288	0.6576	0.6576	nan	0.6576	0.6576
0.0417	0.98	4700	0.0595	0.3405	0.6811	0.6811	nan	0.6811	0.6811
0.1944	1.0	4800	0.0594	0.3346	0.6691	0.6691	nan	0.6691	0.6691

Framework versions

Transformers 4.30.2
Pytorch 2.0.1+cu118
Datasets 2.13.1
Tokenizers 0.13.3

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