Member-only story
Computer Vision, Editorial
Training Faster R-CNN Using TensorFlow’s Object Detection API with a Custom Dataset
Step-by-step tutorial to train a faster R-CNN for object detection with TensorFlow using a custom dataset
Author(s): Buse Yaren Tekin
Recently, object detection has continued to evolve from its current state, and due to its technology, it can be found across almost every technological platform. Whether it is through image classification, recognition, or localization, these are all based on object detection.
Convolutional neural networks (CNNs) can bring together many object recognition and classification techniques together by incorporating deep learning and computer vision methods. In computer vision, convolutional neural networks, as the name suggests, apply a convolution layer in each pixel image in a dataset.
Due to computer vision and deep learning fundamentals in its primary structure, CNNs obtain a different output layer step-by-step by moving the filter we specify onto an image.
“We can build a much brighter future where humans are relieved of menial work using AI capabilities.”
~ Andrew Ng
What is a Faster R-CNN?
Fundamentally, a convolutional neural network performs a pixel-based convolution process on sample images. Faster R-CNN, one of the object recognition algorithms, is one of the R-CNN network types. R-CNN is a regional-based CNN network type.
Over time it has been seen that even R-CNN or even Fast R-CNN has not been enough in terms of performance and accuracy. To ignore the reasons that may negatively affect the performance, a neural network that performs faster is needed and obtained.
