Facebook’s Detectron2 is an object detection platform that is implemented in PyTorch. Its previous version was implemented in Caffe2. It is now more extensible and flexible with high object detection algorithms like DensePose, panoptic feature pyramid networks, and the Mask R-CNN family. It can be trained on single or multiple GPUs. Implementation in PyTorch makes it more reprogrammable and easy to evolve. Read more

October, 2019