Training YOLOv8 with KerasCV on Custom Traffic Light Dataset

[@sudeshnakundu3295]

Great Explanation. Thanks so much

[@LearnOpenCV]

Glad you liked it!

[@alexanderoroudjev4335]

Great video! What would be the next step with the h5 file? I am having issues loading it back up and using it. Could you point to some resources on where I can find this?

[@LearnOpenCV]

You can load the model and run predictions with it. Or you can quantise and optimise the model and deploy it on edge devices. Check how to save and load models here: www.tensorflow.org/tutorials/keras/save_and_load

[@laliborio]

@@LearnOpenCV How do you do to load with the custom_objects from this model? Where can I find the custom_objects so I can load it? Thanks.

[@franciscageorgue2207]

Excelent video! how can I load a image that not exists in the dataset and make a prediction?

[@LearnOpenCV]

Pls visit the blog post where we have shown how to do inference on images after training. Please find the link in the description.

[@coderseparateseason3046]

i use latest keras-cv. i got error. why

[@aduck2619]

Hi! Is it normal to have an ETA/step of 3 hours?

[@LearnOpenCV]

Depending on the system configuration, the model size, input image shape, and the batch size the ETA will vary. Check whether you are training the model on CPU or GPU.

[@nadyasalsabila8835]

I want to experiment yolo v8 with KerasCV by using a dataset that already contains image annotations and stored in txt format, does it have to be converted to XML or can it be used directly? if so, how to do it?

[@LearnOpenCV]

Dataset annotation in any format can be used. However, you will need to modify the dataset preparation code to accommodate for that. In this video and the corresponding article, we have shown how to handle XML annotations.

[@محمدحسنى-ل4ذ]

when i use the code and train the model on 7 epochs , the MaP is always 0.00000e+00 , please give me a sulotion

[@rahulramakrishnan6387]

Hi, are the image normalized as part of the model?

[@LearnOpenCV]

No it is not. This also requires additional research

[@محمدحسنى-ل4ذ]

i run the code on 7 epochs and it always give me MaP : 0.00000e+00