Google to help developers in object identification using TensorFlow Object detection API

June 19, 2017, 3:05 p.m. By: Pranjal Kumar

TensorFlow Object detection API

The new TensorFlow Object detection API by Google will help the developers to identify the objects in the image. The simplicity and improved performance are the two main attraction for this API.

There are various models included in this API with streamlined models. The streamlined models are designed to operate in less complex machines. The API also includes a heavy-duty inception based convolution neural networks.

Google announced the releases of its MobileNets family of lightweight computer vision models, earlier this week. A MobileNet single shot detector comes optimized to run in real-time on a smartphone. These models can handle many tasks like landmark identification, object detection, and facial precogitation.

Developers cannot use their smartphone for the development of software’s that uses object identification technology. This leaves them with only two options either server based setups or large-scale desktops. They can use the cloud services for running the machine learning algorithm but then internet requirement and the latency becomes two major problems. An alternative approach can be to simplify the models.

All the big companies are investing a lot of resource in the mobile models. Recently Facebook announced its Caffe2Go framework for building models to run on the smartphone. This technology was used to develop Facebook’s Style transfer. In the recently conducted I/O conference, Google released TensorFlow Lite which is a streamlined machine learning framework. Apple is also not behind in the competition. In the recently organized WWDC, Apple released CoreML which aims at reducing the difficulty of running machine learning models on the iOS device.

But the Google TensorFlow API has an added advantage of Google’s cloud service. Google wants to make sure that developers feel easy to implement the API. In the same approach, they have released the entire kit pre-packaged with weights and a Jupyter notebook.

The new TensorFlow API will help the developers in creating the application based on the image and facial recognition technology. Now, only time will tell what new features that Google will add in its API.

Image Source: Pjreddie