Microsoft makes image classification model available
Microsoft's release of a pre-trained vision model comes after Google and OpenAI also released open source vision models. The models may not have much enterprise use, though.
The Multimedia Group at Microsoft Bing made a pre-trained vision model, Microsoft Vision Model ResNet-50, freely available to the public.
The machine learning model, trained on millions of images from public data sets, including ImageNet, identifies high-accuracy images from numerous categories, such as household objects, plants and house numbers.
While good for benchmarking and learning, pre-trained image classification models have little practical use for enterprises because the training data sets of images are general, rather than focusing on a specific application, Forrester Research analyst Mike Gualtieri said.
"If your enterprise has the need to classify images by cats, dogs and approximately another 1,000 categories, then it might have some utility," he said. But, for specific use cases, such as screening medical images to detect cancer, such models aren't useful.
Still, the new Microsoft vision tool, available on the Python Package Index as of Feb. 3, shows the tech giant's willingness to provide some of its innovations free to the public.
Amateur developers can practice with these types of models to learn the basics of image classifications. More skilled developers could use Microsoft Vision Model ResNet-50 as a benchmark to compare their own models.
Tech giants
Microsoft's pre-trained vision model joins similar machine learning models from Google and OpenAI, which are also available to download without charge.
In some ways, the big tech and AI vendors are competing both in the commercial arena and in the realm of free and open source software tools available to the public.
If your enterprise has the need to classify images by cats, dogs and approximately another 1,000 categories, then it might have some utility.
Mike GualtieriAnalyst, Forrester Research
These vendors often collaborate with academics on AI research, Gualtieri noted. Those collaborations are likely a key reason why so many AI projects become open source.
Tech giants also choose the open source route with projects to get others to further develop them at no expense to the prominent tech vendors, said Alan Pelz-Sharpe, founder and principal analyst at Deep Analysis.
"Realistically, it's hard to monetize a lot of these open source projects as standalone, but they can be building blocks for much bigger projects and deals," he said.
In fairness, Pelz-Sharpe added, there may also be some altruism there. "It's a good thing to do," he said.
