search

Alzaib / Traffic-Signs-Detection-Tensorflow-YOLOv3-YOLOv4

Final Project for Applications of Machine Learning in Mechatronic Systems
27 stars 14 forks source link

Tensorflow Lite #2

Open HosseinMohammadii opened 3 years ago

HosseinMohammadii commented 3 years ago

Hello. I have read your medium article, nice job. I have 2 question 1) why you didn't use just yolo where it does both detection and classification? 2) Did you convert your final model to tensorflow light? or do you have any experience about converting yolo weights to .tflight file ??

Thanks for your attention.

Alzaib commented 3 years ago

Hi Hossein, 1- This was a course project and as per the requirements, we must have novelty in the project. Therefore we decided to train the YOLO to detect the main class and trained our own custom-built CNN classifier to classify sub-classes detected by the YOLO. As labelling each training image in YOLO format is time-consuming, it would be easier to label the main class in YOLO format, and download images of subclasses from Google/Kaggle. For example, train YOLO to detect animals, and CNN classifier to classify those animals into proper classes (cats, dogs, etc).

2- We did not converted our TF model into TFlite. I do not have any experience with converting yolo weights into tflite .

HosseinMohammadii commented 3 years ago

Thank you for your response