Open-source safety-utcustom-train-SF-RGBD-b5 model - for safety and dangerous area recognition in RGBD images

Safety Utcustom Train SF RGBD B5

Developed by sam1120

A safety vision segmentation model fine-tuned from nvidia/mit-b5, specializing in safety and hazard zone identification in RGBD images

Image Segmentation

Transformers

Open Source License:Other #RGBD safety detection #High-precision segmentation #Industrial scenario applicability

Downloads 17

Release Time : 2/6/2024

Model Overview

This model is an improved version based on the MIT-B5 architecture, specifically designed for processing RGBD image data, effectively identifying safe and hazardous areas in images. It achieves 55.78% accuracy in safety category recognition and up to 99.47% accuracy in hazard category recognition.

Model Features

High-precision hazard zone detection

Achieves 99.47% accuracy and 98.14% intersection over union (IoU) in hazard zone identification

RGBD image processing capability

Specially optimized for processing RGBD image data containing depth information

Balanced performance

Strikes a good balance between overall accuracy (98.18%) and category balance (47.45% IoU for safety categories)

Model Capabilities

Image segmentation

Safety zone identification

Hazard detection

RGBD image processing

Use Cases

Industrial safety monitoring

Real-time monitoring of hazardous areas in factories

Used to monitor hazardous areas in factory environments and provide timely warnings when personnel approach

Can accurately identify over 99% of hazardous areas

Intelligent security systems

Public space safety monitoring

Identifies potential hazard zones in public spaces

Balanced recognition of safe and hazardous areas

🚀 safety-utcustom-train-SF-RGBD-b5

This model is a fine - tuned version of [nvidia/mit - b5](https://huggingface.co/nvidia/mit - b5) on the sam1120/safety - utcustom - TRAIN dataset. It offers high - performance image segmentation capabilities, achieving excellent results in safety - related image analysis.

🚀 Quick Start

This model is ready for use after fine - tuning. You can directly load it for image segmentation tasks on the relevant dataset.

✨ Features

Fine - tuned: Based on the [nvidia/mit - b5](https://huggingface.co/nvidia/mit - b5) model, fine - tuned on the sam1120/safety - utcustom - TRAIN dataset to better adapt to specific safety - related image segmentation tasks.
High accuracy: Achieves high accuracy in multiple evaluation metrics on the evaluation set, such as a low loss value and high mean Iou and accuracy values.

📚 Documentation

Model Evaluation Results

It achieves the following results on the evaluation set:

Loss: 0.0867
Mean Iou: 0.7280
Mean Accuracy: 0.7762
Overall Accuracy: 0.9818
Accuracy Unlabeled: nan
Accuracy Safe: 0.5578
Accuracy Unsafe: 0.9947
Iou Unlabeled: nan
Iou Safe: 0.4745
Iou Unsafe: 0.9814

Training and Evaluation Data

More information needed

Training Procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 4e - 06
train_batch_size: 15
eval_batch_size: 15
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: 120

Training results

Training Loss	Epoch	Step	Accuracy Safe	Accuracy Unlabeled	Accuracy Unsafe	Iou Safe	Iou Unlabeled	Iou Unsafe	Validation Loss	Mean Accuracy	Mean Iou	Overall Accuracy
0.789	0.91	10	0.0203	nan	0.8957	0.0095	0.0	0.8722	0.9555	0.4580	0.2939	0.8698
0.7579	1.82	20	0.0117	nan	0.9614	0.0069	0.0	0.9338	0.8322	0.4866	0.3136	0.9334
0.7103	2.73	30	0.0051	nan	0.9893	0.0043	0.0	0.9604	0.6729	0.4972	0.3216	0.9602
0.676	3.64	40	0.0021	nan	0.9969	0.0020	0.0	0.9675	0.5336	0.4995	0.3232	0.9675
0.5955	4.55	50	0.0001	nan	0.9993	0.0001	0.0	0.9698	0.4440	0.4997	0.3233	0.9698
0.5691	5.45	60	0.0000	nan	0.9997	0.0000	0.0	0.9702	0.3812	0.4999	0.3234	0.9702
0.5067	6.36	70	0.0	nan	0.9996	0.0	0.0	0.9701	0.3590	0.4998	0.3234	0.9701
0.4656	7.27	80	0.0	nan	0.9999	0.0	0.0	0.9703	0.3247	0.4999	0.3234	0.9703
0.4227	8.18	90	0.0	nan	0.9998	0.0	0.0	0.9702	0.3171	0.4999	0.3234	0.9702
0.3898	9.09	100	0.0004	nan	0.9996	0.0004	0.0	0.9701	0.3122	0.5000	0.3235	0.9701
0.3513	10.0	110	0.0	nan	0.9999	0.0	0.0	0.9703	0.2876	0.4999	0.3234	0.9703
0.4157	10.91	120	0.0000	nan	0.9998	0.0000	0.0	0.9703	0.2820	0.4999	0.3234	0.9703
0.3317	11.82	130	0.0	nan	0.9999	0.0	0.0	0.9703	0.2693	0.4999	0.3234	0.9703
0.321	12.73	140	0.0	nan	0.9999	0.0	0.0	0.9704	0.2647	0.4999	0.3235	0.9704
0.2887	13.64	150	0.0	nan	0.9999	0.0	0.0	0.9704	0.2539	0.5000	0.3235	0.9704
0.3008	14.55	160	0.0	nan	0.9999	0.0	0.0	0.9704	0.2536	0.5000	0.3235	0.9704
0.2853	15.45	170	0.0	nan	0.9999	0.0	0.0	0.9704	0.2397	0.5000	0.3235	0.9704
0.2684	16.36	180	0.0	nan	0.9999	0.0	0.0	0.9704	0.2321	0.5000	0.3235	0.9704
0.2585	17.27	190	0.0000	nan	0.9999	0.0000	0.0	0.9704	0.2208	0.5000	0.3235	0.9704
0.2088	18.18	200	0.0084	nan	0.9997	0.0083	0.0	0.9704	0.2011	0.5041	0.3262	0.9704
0.2518	19.09	210	0.0468	nan	0.9989	0.0451	0.0	0.9707	0.2026	0.5228	0.3386	0.9707
0.218	20.0	220	0.0879	nan	0.9984	0.0834	nan	0.9714	0.1889	0.5431	0.5274	0.9715
0.2046	20.91	230	0.1931	nan	0.9969	0.1752	nan	0.9730	0.1847	0.5950	0.5741	0.9732
0.2147	21.82	240	0.2042	nan	0.9968	0.1850	nan	0.9733	0.1766	0.6005	0.5791	0.9734
0.188	22.73	250	0.2020	nan	0.9972	0.1849	nan	0.9735	0.1726	0.5996	0.5792	0.9737
0.2175	23.64	260	0.1898	nan	0.9974	0.1748	nan	0.9734	0.1706	0.5936	0.5741	0.9735
0.2059	24.55	270	0.3006	nan	0.9962	0.2670	nan	0.9754	0.1689	0.6484	0.6212	0.9756
0.1776	25.45	280	0.2870	nan	0.9967	0.2587	nan	0.9755	0.1612	0.6418	0.6171	0.9757
0.1585	26.36	290	0.4254	nan	0.9944	0.3593	nan	0.9773	0.1537	0.7099	0.6683	0.9776
0.1588	27.27	300	0.2798	nan	0.9970	0.2548	nan	0.9756	0.1527	0.6384	0.6152	0.9758
0.153	28.18	310	0.4288	nan	0.9946	0.3646	nan	0.9776	0.1452	0.7117	0.6711	0.9779
0.1623	29.09	320	0.4401	nan	0.9945	0.3726	nan	0.9778	0.1442	0.7173	0.6752	0.9781
0.1603	30.0	330	0.4050	nan	0.9958	0.3562	nan	0.9781	0.1407	0.7004	0.6671	0.9784
0.1694	30.91	340	0.4585	nan	0.9948	0.3911	nan	0.9786	0.1343	0.7266	0.6849	0.9789
0.1585	31.82	350	0.3861	nan	0.9962	0.3433	nan	0.9779	0.1353	0.6912	0.6606	0.9782
0.1342	32.73	360	0.4963	nan	0.9939	0.4132	nan	0.9789	0.1338	0.7451	0.6961	0.9792
0.1358	33.64	370	0.5048	nan	0.9937	0.4182	nan	0.9789	0.1342	0.7493	0.6986	0.9793
0.1493	34.55	380	0.4809	nan	0.9946	0.4080	nan	0.9791	0.1297	0.7377	0.6936	0.9794
0.1435	35.45	390	0.5658	nan	0.9923	0.4518	nan	0.9794	0.1271	0.7791	0.7156	0.9797
0.1305	36.36	400	0.4157	nan	0.9968	0.3758	nan	0.9793	0.1225	0.7062	0.6776	0.9796
0.1496	37.27	410	0.5385	nan	0.9934	0.4420	nan	0.9796	0.1237	0.7659	0.7108	0.9799
0.1445	38.18	420	0.5763	nan	0.9924	0.4615	nan	0.9798	0.1207	0.7843	0.7206	0.9801
0.1307	39.09	430	0.4853	nan	0.9956	0.4244	nan	0.9803	0.1194	0.7404	0.7023	0.9806
0.1379	40.0	440	0.5722	nan	0.9922	0.4557	nan	0.9795	0.1174	0.7822	0.7176	0.9798
0.1202	40.91	450	0.5399	nan	0.9943	0.4544	nan	0.9805	0.1143	0.7671	0.7175	0.9809
0.1239	41.82	460	0.5580	nan	0.9932	0.4558	nan	0.9800	0.1150	0.7756	0.7179	0.9803
0.1183	42.73	470	0.4777	nan	0.9961	0.4236	nan	0.9805	0.1129	0.7369	0.7021	0.9808
0.1202	43.64	480	0.5933	nan	0.9928	0.4793	nan	0.9806	0.1119	0.7930	0.7300	0.9810
0.1276	44.55	490	0.5425	nan	0.9942	0.4561	nan	0.9806	0.1131	0.7683	0.7183	0.9809
0.1172	45.45	500	0.6272	nan	0.9898	0.4700	nan	0.9787	0.1135	0.8085	0.7244	0.9791
0.1288	46.36	510	0.4236	nan	0.9974	0.3898	nan	0.9802	0.1105	0.7105	0.6850	0.9804
0.1185	47.27	520	0.6035	nan	0.9914	0.4711	nan	0.9796	0.1130	0.7975	0.7254	0.9800
0.1045	48.18	530	0.5750	nan	0.9930	0.4679	nan	0.9804	0.1102	0.7840	0.7241	0.9807
0.1211	49.09	540	0.5812	nan	0.9929	0.4715	nan	0.9804	0.1069	0.7870	0.7260	0.9808
0.1206	50.0	550	0.5221	nan	0.9953	0.4528	nan	0.9811	0.1071	0.7587	0.7169	0.9814
0.1193	50.91	560	0.4956	nan	0.9961	0.4398	nan	0.9811	0.1053	0.7459	0.7105	0.9814
0.1116	51.82	570	0.5257	nan	0.9951	0.4528	nan	0.9809	0.1043	0.7604	0.7169	0.9812
0.1218	52.73	580	0.5936	nan	0.9922	0.4724	nan	0.9801	0.1078	0.7929	0.7262	0.9804
0.1284	53.64	590	0.5872	nan	0.9924	0.4696	nan	0.9801	0.1090	0.7907	0.7249	0.9804

📄 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