Our Blogs

Google MobileNets released with Tensorflow computer vision models

June 20, 2017
117 Views
Google MobileNets released with Tensorflow computer vision models
Google has released MobileNets, which is a pre-trained computer vision models for smartphones falling under the Tensorflow Github repository. As a developer, you can choose from a wide range of models that vary based on parameters, accuracy including the time required to process a single image.

While the smallest model has 14 million multiply and adds operations (MACs), the large model consists of 569 MACs. If a model will have to perform advanced computations to accurately predict the image class, the battery power will increase accordingly. If you are a developer, you can exchange accuracy to know the battery power for a particular application.

According to industry analysts, image processing on the smartphone is quicker than image uploading via various Cloud APIs. Moreover, you need not have to worry about the data since they are processed inside the phone. If you look at the cloud platform, you have to be concerned about the privacy of the user. The models are at your disposal free of cost. You can download them and customize as per your requirements.

You should note that the accuracy of the models can be easily compared with the performance of other neural networks. You can determine the accuracy of the models using the LSVRC dataset, which is a large image recognition dataset.

Google MobileNet has predicted five labels for an image. The first label reveals the status of the label. The final label predicts the time interval with which the image label is located on the top five positions of the overall network.