olafenwamoses / firenet Goto Github PK

Request addition to readme of performance metrics of the final model. Based on the model name detection_model-ex-33--loss-4.97.h5 I would guess this is the 33rd iteration of the model?
Cheers!

Hi, i have a queation about how to reduce false positive rate? For in real world applications, there would be many fire like scenarios. I know adding negative samples to the training set may work, but it may reduce recall.

Using the current data set with https://github.com/johnolafenwa/deepstack-trainer results in error No labels found in /content/fire-dataset/train/labels.cache. Can not train without labels I believe due to the dataset being in Pascal VOC format. Deepstack requires yolo formatted data: https://docs.deepstack.cc/custom-models/datasetprep/index.html#change-annotation-to-yolo-format
I uses Roboflow to convert the dataset to yolo format and trained the deepstack model but in production on deepstacl it does not perform well, and I assume there is a data formatting issue: https://github.com/robmarkcole/fire-detection-from-images/blob/master/pytorch/object-detection/yolov5/deepstack/deepstack_custom_training_fire.ipynb

Scope of this request is to update the format, or alternatively provide a second download link in yolo format

Hi, I was about to cite this data set, could you provide the official citation format? Thank a lot.

first thanks for the contribution. I run this project on my computer and it works well.
But as the original video is 102s long, it takes nearly 10 minites to create a tested-video, processing 1267 frames.
So I wonder whether it can be used for real-time fire detection with a camera, and if yes how to make it?

looking forward to your reply,
thank you.

Thank you so much for the great contribution. I'm working on a smart fire fighting sprinkler that would direct water exactly to the burning area after the detection of fire and I'm using raspberry pi for the processing of the frames. I was wondering if this can be deployed on a raspberry pi knowing the huge computational power it requires? Any suggestions? My application should detect fire in real-time.Your help is much appreciated. Thank you in advance

Line number one is following:
from imageai.Detection.Custom import CustomObjectDetection, CustomVideoObjectDetection

But when I try to find the CustomObjectDetection and CustomVideoObjectDetection files in Custom folder which is inside the Detection folder, I am unable to locate them and my code is not able to import the same.

Please help me with this issue.

Dear FireNet Contributors,

I hope this message finds you well. I am reaching out to discuss potential enhancements to the FireNet project, an initiative that I hold in high regard for its commitment to leveraging artificial intelligence in the service of public safety.

Upon reviewing the current state of the project, I have identified a few areas where I believe we could introduce improvements that would significantly augment the system's fire detection capabilities. Below, I outline these suggestions and provide a rationale for each.

1. Integration of Thermal Imaging Data

   • Rationale: Traditional visual spectrum cameras can be limited in smoke-filled or low-visibility environments. By incorporating thermal imaging, FireNet could detect heat signatures associated with fires even in challenging conditions, thus improving detection reliability.

2. Real-Time Data Analysis for Faster Response

   • Rationale: The current model may benefit from a more streamlined data processing pipeline. Implementing a real-time analysis framework could reduce latency and enable quicker response times, which is critical in emergency situations.

3. Enhanced Model Training with Diverse Datasets

   • Rationale: While the current dataset is a robust starting point, expanding it to include a wider variety of fire scenarios, including different lighting and weather conditions, could improve the model's accuracy and generalisation capabilities.

4. Deployment of Edge Computing for Localised Processing

   • Rationale: Edge computing can facilitate faster processing by bringing computation closer to the data source. This would be particularly beneficial for remote or bandwidth-limited locations where sending data to the cloud is not feasible.

5. Utilisation of an Ensemble of Models for Improved Accuracy

   • Rationale: An ensemble approach, where multiple models are used in conjunction, can often yield better results than any single model. This could help in reducing false positives and increasing the confidence of fire detection.

6. Community Engagement for Continuous Improvement

   • Rationale: Encouraging the community to participate in the project by sharing their own datasets and detection scenarios can lead to a more robust and tested system. This collaborative approach can accelerate innovation and refinement of the FireNet project.

I am eager to hear your thoughts on these suggestions and would be delighted to contribute further to the discussion. The FireNet project has the potential to make a significant impact on fire safety, and I believe that with these enhancements, we can take a substantial leap forward in its development.

Thank you for your time and consideration.

Best regards,
yihong1120

Hi @OlafenwaMoses , could you kindly opensource the dataset used for training the model please.

I cannot get the fire_net.py to work. some package is always the wrong version, causing some missing function or module.
So can you specify a working combination of versions of these packages

你好，我开源了1个火灾检测数据集。数据是我自己爬虫得到的，并采用labelimg标注。目前想将您的数据合并到我的数据集上，我会在github上引用您的链接，请问可以么？
Hello, I open sourced a fire detection data set. The data was obtained by my own crawler and marked with labelimg. I currently want to merge your data into my dataset, I will Cite your link on github, is it okay?

https://github.com/gengyanlei/fire-detect-yolov4

Hello,
thank you very much for contribution
how can i use this in real time videos for fire detection?

And also, how can i train and model the scenario and how ı run your model yolov3?

is there any manual for installation, modelling and real time detection how we will do?

[email protected] please send this here. best regards.

how to test the program?

I run with the command

python3 fire_net.py

it showed illegal instruction(core dumped).

Does my Computer doesn't support or doesn't have any configuration?

Hey, i have an issue of using for real time detection. Can u plz share a code for RT Detection?
[email protected]

First of all, I want to thank you for sharing this code. When I compile the code, I face 2 errors in detector.loadModel()

The main function is this :
detection_model_json = json.load(open(self._detection_config_json_path))
Lines 911 and 642 in init.py file

what is the reason for this error?

I changed the lines in detect_from_video() function like this :

detector.setModelPath(detection_model_path=os.path.join("F:\PROJECTS\fire detection python\FireNET-master\pretrained-yolov3.h5", "detection_model-ex-33--loss-4.97.h5"))

detector.setJsonPath(configuration_json=os.path.join("F:\PROJECTS\fire detection python\FireNET-master\detection_config.json", "detection_config.json"))

raise RuntimeError("You are trying to use a Tensorflow model with ImageAI. ImageAI now uses PyTorch as backed as from version 3.0.2 . If you want to use the Tensorflow models or a customly trained '.h5' model, install ImageAI 2.1.6 or earlier. To use the latest Pytorch models, see the documentation in https://imageai.readthedocs.io/")

Can you please make sure your code works with newer Python and ImageAI version avaiable now?

Hi

thanks for sharing. i have perform a training using the code (with the directory change to reflect the actual position of the dataset) but i encountered..

Please use Model.fit, which supports generators.
WARNING:tensorflow:Model failed to serialize as JSON. Ignoring... Layer YoloLayer has arguments in __init__ and therefore must override get_config.
Epoch 1/100
....

Error encountered after execution:

ValueError: tf.function-decorated function tried to create variables on non-first call.

i have raised the same query on imageai, if it is answered there, will close the thread here. thank you