gaiasd / dfiredataset Goto Github PK

View Code? Open in Web Editor NEW

165.0 5.0 25.0 65 KB

D-Fire: an image data set for fire and smoke detection.

License: Other

Python 100.00%

computer-vision dataset fire-detection yolo smoke-detection

dfiredataset's Introduction

D-Fire: an image dataset for fire and smoke detection

Authors: Researchers from Gaia, solutions on demand (GAIA)

About

D-Fire is an image dataset of fire and smoke occurrences designed for machine learning and object detection algorithms with more than 21,000 images.

Number of images

Number of bounding boxes

Category	# Images
Only fire	1,164
Only smoke	5,867
Fire and smoke	4,658
None	9,838

Class	# Bounding boxes
Fire	14,692
Smoke	11,865

All images were annotated according to the YOLO format (normalized coordinates between 0 and 1). However, we provide the yolo2pixel function that converts coordinates in YOLO format to coordinates in pixels.

Examples

Download

D-Fire dataset (only images and labels).
Training, validation and test sets.
Some surveillance videos.
Some models trained with the D-Fire dataset.
For more surveillance videos, request your registration on our environmental monitoring website "Apaga o Fogo!" (Put out the Fire!).

Citation

Please cite the following paper if you use our image database:

Pedro Vinícius Almeida Borges de Venâncio, Adriano Chaves Lisboa, Adriano Vilela Barbosa: An automatic fire detection system based on deep convolutional neural networks for low-power, resource-constrained devices. In: Neural Computing and Applications, 2022.

If you use our surveillance videos, please cite the following paper:

Pedro Vinícius Almeida Borges de Venâncio, Roger Júnio Campos, Tamires Martins Rezende, Adriano Chaves Lisboa, Adriano Vilela Barbosa: A hybrid method for fire detection based on spatial and temporal patterns. In: Neural Computing and Applications, 2023.

dfiredataset's People

Contributors

Stargazers

Watchers

dfiredataset's Issues

Enhancing D-Fire Dataset's Applicability for Diverse Fire Detection Scenarios

Dear D-Fire Dataset Contributors,

I trust this message finds you in good health and high spirits. I am writing to you today to address a matter of great importance concerning the D-Fire image dataset, which has proven to be an invaluable resource for the development of fire and smoke detection algorithms.

Having delved into the dataset, I have observed its robustness and the meticulous effort that has gone into its curation. However, I believe that there is an opportunity to further enhance its utility by considering the following suggestions:

Diversity in Fire Contexts: The inclusion of fire images from a wider array of contexts, such as forest fires at different times of the day and urban fires in various architectural settings, could significantly improve the dataset's comprehensiveness.
Varied Lighting Conditions: Fire and smoke detection in low-light or night-time scenarios can be particularly challenging. Augmenting the dataset with images captured under these conditions would be beneficial for developing more resilient detection models.
Annotation Refinement: While the current YOLO format annotations are quite useful, providing additional formats such as Pascal VOC or COCO could facilitate the use of the dataset across different object detection frameworks.
Temporal Data Annotation: For the surveillance videos, annotations that include temporal information could enable the development of models that leverage temporal dynamics for improved detection accuracy.
Live Data Stream Integration: Establishing a protocol for integrating live data streams from surveillance cameras could pave the way for real-time fire detection and the development of systems that learn continuously from evolving data.

I am keen to hear your thoughts on these propositions and to explore potential collaborations to implement these enhancements. By addressing these aspects, I am confident that we can elevate the D-Fire dataset to new heights, making it even more versatile for researchers and practitioners in the field of fire detection.

Thank you for your time and consideration. I eagerly await your response and am excited about the prospect of contributing to the evolution of the D-Fire dataset.

Best regards,
yihong1120

Thanks to the author's hard work, thank the author is willing to disclose the data set.I want to test the accuracy of my model. Can I provide a verification set?

Thanks to the author's hard work, thank the author is willing to disclose the data set!
I download dataset,it contain 4 folds :
CAD FP PublicDataset WEB
In your paper, it is said that the test set contains 4306 pictures, but in each folder does not contain 4306*2 pictures, and the number of pictures in the folder does not match (4306+17221) *2 pictures。
So I speculate that you mix the picture inside multiple folders，I want to test the accuracy of my model. Can you provide a test set?

can you show me your train val test sets?

Thanks to the your hard work!
I'm sorry I don't konw the relationship between the four folders. The name has the differences like noise and mirror(data agumentation), but some name like "FacultyofElectricalEngineeringSplitUniversity_WEBSmoke1869.jpg" I can't understand. So can you explian it and give me your train, val, test sets according to the table that introduces categories. I can't summary the images as your category.
My english is not good. Hope you can understand my intension. I will cite your paper in your publications if I use this dataset.

Pre-trained smoke detection models?

Intersting repo. Your sample image shows smoke + fire detection, but there is no pre-trained models available. Can you please share the that model produced that? I'm specifically interested in smoke part.

Regarding data labels

Hi,

Thanks for making the dataset public. I also read the Medium post, it looks very promising.
I just downloaded the dataset, but the labels for both train and test are all blank. Can you confirm if this is a glitch or am I missing something here.
Looking forward to your reply.
Best Regards,
Pramit