What YOLO algorithm can I use for images with noise as I will implement it in real time?

Question

I want to detect drivers with or without seatbelts at crossroads. For that, as it is real-time, I am going to use the YOLO algorithm/model. For training data sets (the images) I need to collect, I placed a camera. By recording it and collecting images from there, I am getting images with more noise.

Can I use these images for training? Also, which YOLO version should I use? What are the important points that I should consider for training datasets?

I want to use any version of YOLO compatible with TensorFlow.

Answer 1

It is much better to know basic mechanics of convnets first ,rather than diving straight into complicated models .

For training data sets (the images) I need to collect, I placed a camera. By recording it and collecting images from there, I am getting images with more noise. Can I use these images for training? Also, which yolo version should I use? What are the important points that I should consider for training datasets?

After you are good with the theory part most of your questions will be answered , otherwise you would endup with nothing but buzzwords.

I want to use any version of yolo compatible with tensorflow.

Tensorflow is a framework for building neural networks , so in theory you can build any network with it so compatibility is not at all a problem.

Answer 2

As long as all training samples have same type and level of noise, it shouldn't affect the outcome/accuracy.

As far as real time performance is concerned, you should compare the "Params, FLops and Inference time @ B1 (Batch size 1)" numbers of different yolo versions. Lower the number, faster the inference.

You can start by exploring yolov5 and Yolo-Fastestv2 at the links below:

https://github.com/ultralytics/yolov5

https://github.com/dog-qiuqiu/Yolo-FastestV2

Unfortunately most of the implementations are in PyTorch, but you can easily convert a PyTorch model to a Tensorflow one by first converting it to an ONNX format.