|Facebook Releases Detectron2|
|Written by Kay Ewbank|
|Tuesday, 05 November 2019|
Facebook AI Research has released a new version of its Detectron software that implements object-detection algorithms. The new version is powered by the PyTorch deep learning framework, and has new features including panoptic segmentation, densepose, Cascade R-CNN, and rotated bounding boxes.
Facebook AI Research is using Detectron2 to design and train the next-generation pose detection models that power Smart Camera, the AI camera system in Facebook’s Portal video-calling devices.
The developers say the new release has been rewritten from the ground up. It originates from maskrcnn-benchmark, which is a convolutional neural network that will place a mask around objects recognized in the image. Detectron is based on a number of types of neural networks and it is written in Python and uses the Caffe2 deep learning library. The new version provides fast training on single or multiple GPU servers.
The new version has a more modular design and is designed to be extensible. The new design means users can plug custom module implementations into almost any part of an object detection system. The developers say this means that new research projects can be written in hundreds of lines of code with a clean separation between the core Detectron2 library and the novel research implementation.
Object-detection algorithms supported by the new version include DensePose and panoptic feature pyramid networks. DensePose (dense human pose) estimation aims at mapping all human pixels of an RGB image to the 3D surface of the human body. Panoptic segmentation combines instance segmentation which is used to recognize distinct foreground objects such as animals or people with the semantic segmentation which labels pixels in the image background with classes, such as road, sky, or grass.
or email your comment to: firstname.lastname@example.org
|Last Updated ( Tuesday, 05 November 2019 )|