|Facebook Open Sources Detectron Object Detection|
|Written by Alex Armsrtong|
|Tuesday, 23 January 2018|
The way big companies are open sourcing significant AI is both gratifying and slightly worrying. AI is the biggest revolution since we discovered fire and started making tools. FaceBook AI Research has added to the list of what is available by open sourcing its Detectron project.
Object detection was until recently a cutting edge research endeavour that mostly didn't work well enough for any real applications. Today we have a range of techniques, some, but not all, based on neural networks, that are good enough to use as the basis of other programs. Facebook's Detectron project has been the basis for many other of its AI projects and now you can download and use the code under an Apache 2.0 licence, so no patent worries, from GitHub. You do need a machine that has a GPU, NVIDIA CUDA, for it to work, however.
"These algorithms, powered by Detectron, provide intuitive models for important computer vision tasks, such as instance segmentation, and have played a key role in the unprecedented advancement of visual perception systems that our community has achieved in recent years.
Beyond research, a number of Facebook teams use this platform to train custom models for a variety of applications including augmented reality and community integrity. Once trained, these models can be deployed in the cloud and on mobile devices, powered by the highly efficient Caffe2 runtime."
It is based on a number of types of neural networks and it is written in Python and uses the Caffe2 deep learning library. It includes an implementation of Mask R-CNN, a convolutional neural network that will place a mask around objects recognized in the image. It is currently regarded as state of the art, winning the Marr Prize in 2017. It also implements five other very good networks just in case you want to experiment.
Given all of the open source software from Google and others, we almost have difficulty choosing what to use, but in most cases the hardware demands are high and this is still a limiting factor in what you can invent. You can't, for example, run any of this on, say, a Raspberry Pi. You are going to need a cloud-based service for most practical applications.
or email your comment to: firstname.lastname@example.org
|Last Updated ( Tuesday, 23 January 2018 )|