The reason I choose SSD MobileNet v2 320x320 model is quite similar as the blog post. (And the performance of those models already have been proved!) I wanted to save my time on implementing object detection model on my own. When it comes to training the model, I followed steps described in Tensorflow Blog. Training and Validating Choosing the model Width and height column means the size of image and xmin, xmax, ymin, ymax columns represents the bounding box of detected object. (plus, splitting the training and test set) Next thing what we need to do is parsing the annotation xml file and generating csv file with label information. The below is how annotation file looks like. The 'Bikes Helmet Dataset' includes 764 images and annotations in PascalVOC format. It includes 764 images and annotations in PascalVOC format.Īccordingly, I decided to solve this problem by object detection approach. So here, I decided to choose the approach depending on which dataset I could be available of.Īfter few hours of googling, luckily, I could find the 'Bikes Helmet Dataset' out. I would rather wanted to find prepared dataset which is good to go. I didn't want to spend too much time on gathering and cleaning dataset. Since this project has begun with my personal curiosity, We could get to the goal with object detection as well. When we have raw pixel image as input and label of each image.
It could be achieved by several different approaches.įor example, binary classification can be used to classify the image(wearing a helmet or not) What I wanted to do with this application was to classify the images by if people wearing bike helmet or not. Then developed an mobile application with react-native which can inference real-time. In here, I used Bikes Helmet Dataset and SSD MobileNet v2 320x320 model(provided by Tensorflow Object Detection API).įirst, I trained the model in Python using Google Colab, The problem I wanted to solve was to check whether person wearing helmet with mobile application. How I deployed 'Helmet Detection' model with React Native application Object Detection, Deep Learning, Tensorflow.js, React Native Deploying Deep Learning model with React Native
0 kommentar(er)
